2 * Copyright (c) 2002 Luigi Rizzo, Universita` di Pisa
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
7 * 1. Redistributions of source code must retain the above copyright
8 * notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 * notice, this list of conditions and the following disclaimer in the
11 * documentation and/or other materials provided with the distribution.
13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * Implement IP packet firewall (new version)
35 #if !defined(KLD_MODULE)
40 #error IPFIREWALL requires INET.
43 #include "opt_inet6.h"
44 #include "opt_ipsec.h"
46 #include <sys/param.h>
47 #include <sys/systm.h>
48 #include <sys/condvar.h>
49 #include <sys/malloc.h>
51 #include <sys/kernel.h>
54 #include <sys/module.h>
56 #include <sys/rwlock.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/syslog.h>
61 #include <sys/ucred.h>
63 #include <net/radix.h>
64 #include <net/route.h>
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_pcb.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/ip_icmp.h>
72 #include <netinet/ip_fw.h>
73 #include <netinet/ip_divert.h>
74 #include <netinet/ip_dummynet.h>
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_timer.h>
77 #include <netinet/tcp_var.h>
78 #include <netinet/tcpip.h>
79 #include <netinet/udp.h>
80 #include <netinet/udp_var.h>
82 #include <netgraph/ng_ipfw.h>
84 #include <altq/if_altq.h>
87 #include <netinet6/ipsec.h>
90 #include <netinet/ip6.h>
91 #include <netinet/icmp6.h>
93 #include <netinet6/scope6_var.h>
96 #include <netinet/if_ether.h> /* XXX for ETHERTYPE_IP */
98 #include <machine/in_cksum.h> /* XXX for in_cksum */
101 * set_disable contains one bit per set value (0..31).
102 * If the bit is set, all rules with the corresponding set
103 * are disabled. Set RESVD_SET(31) is reserved for the default rule
104 * and rules that are not deleted by the flush command,
105 * and CANNOT be disabled.
106 * Rules in set RESVD_SET can only be deleted explicitly.
108 static u_int32_t set_disable;
110 static int fw_verbose;
111 static int verbose_limit;
113 static struct callout ipfw_timeout;
114 static uma_zone_t ipfw_dyn_rule_zone;
115 #define IPFW_DEFAULT_RULE 65535
118 * Data structure to cache our ucred related
119 * information. This structure only gets used if
120 * the user specified UID/GID based constraints in
124 gid_t fw_groups[NGROUPS];
130 #define IPFW_TABLES_MAX 128
132 struct ip_fw *rules; /* list of rules */
133 struct ip_fw *reap; /* list of rules to reap */
134 struct radix_node_head *tables[IPFW_TABLES_MAX];
137 #define IPFW_LOCK_INIT(_chain) \
138 rw_init(&(_chain)->rwmtx, "IPFW static rules")
139 #define IPFW_LOCK_DESTROY(_chain) rw_destroy(&(_chain)->rwmtx)
140 #define IPFW_WLOCK_ASSERT(_chain) do { \
141 rw_assert(&(_chain)->rwmtx, RA_WLOCKED); \
142 NET_ASSERT_GIANT(); \
145 #define IPFW_RLOCK(p) rw_rlock(&(p)->rwmtx)
146 #define IPFW_RUNLOCK(p) rw_runlock(&(p)->rwmtx)
147 #define IPFW_WLOCK(p) rw_wlock(&(p)->rwmtx)
148 #define IPFW_WUNLOCK(p) rw_wunlock(&(p)->rwmtx)
151 * list of rules for layer 3
153 static struct ip_fw_chain layer3_chain;
155 MALLOC_DEFINE(M_IPFW, "IpFw/IpAcct", "IpFw/IpAcct chain's");
156 MALLOC_DEFINE(M_IPFW_TBL, "ipfw_tbl", "IpFw tables");
159 struct radix_node rn[2];
160 struct sockaddr_in addr, mask;
164 static int fw_debug = 1;
165 static int autoinc_step = 100; /* bounded to 1..1000 in add_rule() */
167 extern int ipfw_chg_hook(SYSCTL_HANDLER_ARGS);
170 SYSCTL_NODE(_net_inet_ip, OID_AUTO, fw, CTLFLAG_RW, 0, "Firewall");
171 SYSCTL_PROC(_net_inet_ip_fw, OID_AUTO, enable,
172 CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3, &fw_enable, 0,
173 ipfw_chg_hook, "I", "Enable ipfw");
174 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, autoinc_step, CTLFLAG_RW,
175 &autoinc_step, 0, "Rule number autincrement step");
176 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, one_pass,
177 CTLFLAG_RW | CTLFLAG_SECURE3,
179 "Only do a single pass through ipfw when using dummynet(4)");
180 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, debug, CTLFLAG_RW,
181 &fw_debug, 0, "Enable printing of debug ip_fw statements");
182 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose,
183 CTLFLAG_RW | CTLFLAG_SECURE3,
184 &fw_verbose, 0, "Log matches to ipfw rules");
185 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, verbose_limit, CTLFLAG_RW,
186 &verbose_limit, 0, "Set upper limit of matches of ipfw rules logged");
189 * Description of dynamic rules.
191 * Dynamic rules are stored in lists accessed through a hash table
192 * (ipfw_dyn_v) whose size is curr_dyn_buckets. This value can
193 * be modified through the sysctl variable dyn_buckets which is
194 * updated when the table becomes empty.
196 * XXX currently there is only one list, ipfw_dyn.
198 * When a packet is received, its address fields are first masked
199 * with the mask defined for the rule, then hashed, then matched
200 * against the entries in the corresponding list.
201 * Dynamic rules can be used for different purposes:
203 * + enforcing limits on the number of sessions;
204 * + in-kernel NAT (not implemented yet)
206 * The lifetime of dynamic rules is regulated by dyn_*_lifetime,
207 * measured in seconds and depending on the flags.
209 * The total number of dynamic rules is stored in dyn_count.
210 * The max number of dynamic rules is dyn_max. When we reach
211 * the maximum number of rules we do not create anymore. This is
212 * done to avoid consuming too much memory, but also too much
213 * time when searching on each packet (ideally, we should try instead
214 * to put a limit on the length of the list on each bucket...).
216 * Each dynamic rule holds a pointer to the parent ipfw rule so
217 * we know what action to perform. Dynamic rules are removed when
218 * the parent rule is deleted. XXX we should make them survive.
220 * There are some limitations with dynamic rules -- we do not
221 * obey the 'randomized match', and we do not do multiple
222 * passes through the firewall. XXX check the latter!!!
224 static ipfw_dyn_rule **ipfw_dyn_v = NULL;
225 static u_int32_t dyn_buckets = 256; /* must be power of 2 */
226 static u_int32_t curr_dyn_buckets = 256; /* must be power of 2 */
228 static struct mtx ipfw_dyn_mtx; /* mutex guarding dynamic rules */
229 #define IPFW_DYN_LOCK_INIT() \
230 mtx_init(&ipfw_dyn_mtx, "IPFW dynamic rules", NULL, MTX_DEF)
231 #define IPFW_DYN_LOCK_DESTROY() mtx_destroy(&ipfw_dyn_mtx)
232 #define IPFW_DYN_LOCK() mtx_lock(&ipfw_dyn_mtx)
233 #define IPFW_DYN_UNLOCK() mtx_unlock(&ipfw_dyn_mtx)
234 #define IPFW_DYN_LOCK_ASSERT() mtx_assert(&ipfw_dyn_mtx, MA_OWNED)
237 * Timeouts for various events in handing dynamic rules.
239 static u_int32_t dyn_ack_lifetime = 300;
240 static u_int32_t dyn_syn_lifetime = 20;
241 static u_int32_t dyn_fin_lifetime = 1;
242 static u_int32_t dyn_rst_lifetime = 1;
243 static u_int32_t dyn_udp_lifetime = 10;
244 static u_int32_t dyn_short_lifetime = 5;
247 * Keepalives are sent if dyn_keepalive is set. They are sent every
248 * dyn_keepalive_period seconds, in the last dyn_keepalive_interval
249 * seconds of lifetime of a rule.
250 * dyn_rst_lifetime and dyn_fin_lifetime should be strictly lower
251 * than dyn_keepalive_period.
254 static u_int32_t dyn_keepalive_interval = 20;
255 static u_int32_t dyn_keepalive_period = 5;
256 static u_int32_t dyn_keepalive = 1; /* do send keepalives */
258 static u_int32_t static_count; /* # of static rules */
259 static u_int32_t static_len; /* size in bytes of static rules */
260 static u_int32_t dyn_count; /* # of dynamic rules */
261 static u_int32_t dyn_max = 4096; /* max # of dynamic rules */
263 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_buckets, CTLFLAG_RW,
264 &dyn_buckets, 0, "Number of dyn. buckets");
265 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, curr_dyn_buckets, CTLFLAG_RD,
266 &curr_dyn_buckets, 0, "Current Number of dyn. buckets");
267 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_count, CTLFLAG_RD,
268 &dyn_count, 0, "Number of dyn. rules");
269 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_max, CTLFLAG_RW,
270 &dyn_max, 0, "Max number of dyn. rules");
271 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, static_count, CTLFLAG_RD,
272 &static_count, 0, "Number of static rules");
273 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_ack_lifetime, CTLFLAG_RW,
274 &dyn_ack_lifetime, 0, "Lifetime of dyn. rules for acks");
275 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_syn_lifetime, CTLFLAG_RW,
276 &dyn_syn_lifetime, 0, "Lifetime of dyn. rules for syn");
277 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_fin_lifetime, CTLFLAG_RW,
278 &dyn_fin_lifetime, 0, "Lifetime of dyn. rules for fin");
279 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_rst_lifetime, CTLFLAG_RW,
280 &dyn_rst_lifetime, 0, "Lifetime of dyn. rules for rst");
281 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_udp_lifetime, CTLFLAG_RW,
282 &dyn_udp_lifetime, 0, "Lifetime of dyn. rules for UDP");
283 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_short_lifetime, CTLFLAG_RW,
284 &dyn_short_lifetime, 0, "Lifetime of dyn. rules for other situations");
285 SYSCTL_INT(_net_inet_ip_fw, OID_AUTO, dyn_keepalive, CTLFLAG_RW,
286 &dyn_keepalive, 0, "Enable keepalives for dyn. rules");
290 * IPv6 specific variables
292 SYSCTL_DECL(_net_inet6_ip6);
294 static struct sysctl_ctx_list ip6_fw_sysctl_ctx;
295 static struct sysctl_oid *ip6_fw_sysctl_tree;
297 #endif /* SYSCTL_NODE */
299 static int fw_deny_unknown_exthdrs = 1;
303 * L3HDR maps an ipv4 pointer into a layer3 header pointer of type T
304 * Other macros just cast void * into the appropriate type
306 #define L3HDR(T, ip) ((T *)((u_int32_t *)(ip) + (ip)->ip_hl))
307 #define TCP(p) ((struct tcphdr *)(p))
308 #define UDP(p) ((struct udphdr *)(p))
309 #define ICMP(p) ((struct icmphdr *)(p))
310 #define ICMP6(p) ((struct icmp6_hdr *)(p))
313 icmptype_match(struct icmphdr *icmp, ipfw_insn_u32 *cmd)
315 int type = icmp->icmp_type;
317 return (type <= ICMP_MAXTYPE && (cmd->d[0] & (1<<type)) );
320 #define TT ( (1 << ICMP_ECHO) | (1 << ICMP_ROUTERSOLICIT) | \
321 (1 << ICMP_TSTAMP) | (1 << ICMP_IREQ) | (1 << ICMP_MASKREQ) )
324 is_icmp_query(struct icmphdr *icmp)
326 int type = icmp->icmp_type;
328 return (type <= ICMP_MAXTYPE && (TT & (1<<type)) );
333 * The following checks use two arrays of 8 or 16 bits to store the
334 * bits that we want set or clear, respectively. They are in the
335 * low and high half of cmd->arg1 or cmd->d[0].
337 * We scan options and store the bits we find set. We succeed if
339 * (want_set & ~bits) == 0 && (want_clear & ~bits) == want_clear
341 * The code is sometimes optimized not to store additional variables.
345 flags_match(ipfw_insn *cmd, u_int8_t bits)
350 if ( ((cmd->arg1 & 0xff) & bits) != 0)
351 return 0; /* some bits we want set were clear */
352 want_clear = (cmd->arg1 >> 8) & 0xff;
353 if ( (want_clear & bits) != want_clear)
354 return 0; /* some bits we want clear were set */
359 ipopts_match(struct ip *ip, ipfw_insn *cmd)
361 int optlen, bits = 0;
362 u_char *cp = (u_char *)(ip + 1);
363 int x = (ip->ip_hl << 2) - sizeof (struct ip);
365 for (; x > 0; x -= optlen, cp += optlen) {
366 int opt = cp[IPOPT_OPTVAL];
368 if (opt == IPOPT_EOL)
370 if (opt == IPOPT_NOP)
373 optlen = cp[IPOPT_OLEN];
374 if (optlen <= 0 || optlen > x)
375 return 0; /* invalid or truncated */
383 bits |= IP_FW_IPOPT_LSRR;
387 bits |= IP_FW_IPOPT_SSRR;
391 bits |= IP_FW_IPOPT_RR;
395 bits |= IP_FW_IPOPT_TS;
399 return (flags_match(cmd, bits));
403 tcpopts_match(struct tcphdr *tcp, ipfw_insn *cmd)
405 int optlen, bits = 0;
406 u_char *cp = (u_char *)(tcp + 1);
407 int x = (tcp->th_off << 2) - sizeof(struct tcphdr);
409 for (; x > 0; x -= optlen, cp += optlen) {
411 if (opt == TCPOPT_EOL)
413 if (opt == TCPOPT_NOP)
427 bits |= IP_FW_TCPOPT_MSS;
431 bits |= IP_FW_TCPOPT_WINDOW;
434 case TCPOPT_SACK_PERMITTED:
436 bits |= IP_FW_TCPOPT_SACK;
439 case TCPOPT_TIMESTAMP:
440 bits |= IP_FW_TCPOPT_TS;
445 return (flags_match(cmd, bits));
449 iface_match(struct ifnet *ifp, ipfw_insn_if *cmd)
451 if (ifp == NULL) /* no iface with this packet, match fails */
453 /* Check by name or by IP address */
454 if (cmd->name[0] != '\0') { /* match by name */
457 if (fnmatch(cmd->name, ifp->if_xname, 0) == 0)
460 if (strncmp(ifp->if_xname, cmd->name, IFNAMSIZ) == 0)
467 TAILQ_FOREACH(ia, &ifp->if_addrhead, ifa_link) {
468 if (ia->ifa_addr == NULL)
470 if (ia->ifa_addr->sa_family != AF_INET)
472 if (cmd->p.ip.s_addr == ((struct sockaddr_in *)
473 (ia->ifa_addr))->sin_addr.s_addr)
474 return(1); /* match */
477 return(0); /* no match, fail ... */
481 * The verify_path function checks if a route to the src exists and
482 * if it is reachable via ifp (when provided).
484 * The 'verrevpath' option checks that the interface that an IP packet
485 * arrives on is the same interface that traffic destined for the
486 * packet's source address would be routed out of. The 'versrcreach'
487 * option just checks that the source address is reachable via any route
488 * (except default) in the routing table. These two are a measure to block
489 * forged packets. This is also commonly known as "anti-spoofing" or Unicast
490 * Reverse Path Forwarding (Unicast RFP) in Cisco-ese. The name of the knobs
491 * is purposely reminiscent of the Cisco IOS command,
493 * ip verify unicast reverse-path
494 * ip verify unicast source reachable-via any
496 * which implements the same functionality. But note that syntax is
497 * misleading. The check may be performed on all IP packets whether unicast,
498 * multicast, or broadcast.
501 verify_path(struct in_addr src, struct ifnet *ifp)
504 struct sockaddr_in *dst;
506 bzero(&ro, sizeof(ro));
508 dst = (struct sockaddr_in *)&(ro.ro_dst);
509 dst->sin_family = AF_INET;
510 dst->sin_len = sizeof(*dst);
512 rtalloc_ign(&ro, RTF_CLONING);
514 if (ro.ro_rt == NULL)
518 * If ifp is provided, check for equality with rtentry.
519 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
520 * in order to pass packets injected back by if_simloop():
521 * if useloopback == 1 routing entry (via lo0) for our own address
522 * may exist, so we need to handle routing assymetry.
524 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
529 /* if no ifp provided, check if rtentry is not default route */
531 satosin(rt_key(ro.ro_rt))->sin_addr.s_addr == INADDR_ANY) {
536 /* or if this is a blackhole/reject route */
537 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
542 /* found valid route */
549 * ipv6 specific rules here...
552 icmp6type_match (int type, ipfw_insn_u32 *cmd)
554 return (type <= ICMP6_MAXTYPE && (cmd->d[type/32] & (1<<(type%32)) ) );
558 flow6id_match( int curr_flow, ipfw_insn_u32 *cmd )
561 for (i=0; i <= cmd->o.arg1; ++i )
562 if (curr_flow == cmd->d[i] )
567 /* support for IP6_*_ME opcodes */
569 search_ip6_addr_net (struct in6_addr * ip6_addr)
573 struct in6_ifaddr *fdm;
574 struct in6_addr copia;
576 TAILQ_FOREACH(mdc, &ifnet, if_link)
577 for (mdc2 = mdc->if_addrlist.tqh_first; mdc2;
578 mdc2 = mdc2->ifa_list.tqe_next) {
581 if (mdc2->ifa_addr->sa_family == AF_INET6) {
582 fdm = (struct in6_ifaddr *)mdc2;
583 copia = fdm->ia_addr.sin6_addr;
584 /* need for leaving scope_id in the sock_addr */
585 in6_clearscope(&copia);
586 if (IN6_ARE_ADDR_EQUAL(ip6_addr, &copia))
594 verify_path6(struct in6_addr *src, struct ifnet *ifp)
597 struct sockaddr_in6 *dst;
599 bzero(&ro, sizeof(ro));
601 dst = (struct sockaddr_in6 * )&(ro.ro_dst);
602 dst->sin6_family = AF_INET6;
603 dst->sin6_len = sizeof(*dst);
604 dst->sin6_addr = *src;
605 rtalloc_ign((struct route *)&ro, RTF_CLONING);
607 if (ro.ro_rt == NULL)
611 * if ifp is provided, check for equality with rtentry
612 * We should use rt->rt_ifa->ifa_ifp, instead of rt->rt_ifp,
613 * to support the case of sending packets to an address of our own.
614 * (where the former interface is the first argument of if_simloop()
615 * (=ifp), the latter is lo0)
617 if (ifp != NULL && ro.ro_rt->rt_ifa->ifa_ifp != ifp) {
622 /* if no ifp provided, check if rtentry is not default route */
624 IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(ro.ro_rt))->sin6_addr)) {
629 /* or if this is a blackhole/reject route */
630 if (ifp == NULL && ro.ro_rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
635 /* found valid route */
641 hash_packet6(struct ipfw_flow_id *id)
644 i = (id->dst_ip6.__u6_addr.__u6_addr32[2]) ^
645 (id->dst_ip6.__u6_addr.__u6_addr32[3]) ^
646 (id->src_ip6.__u6_addr.__u6_addr32[2]) ^
647 (id->src_ip6.__u6_addr.__u6_addr32[3]) ^
648 (id->dst_port) ^ (id->src_port);
653 is_icmp6_query(int icmp6_type)
655 if ((icmp6_type <= ICMP6_MAXTYPE) &&
656 (icmp6_type == ICMP6_ECHO_REQUEST ||
657 icmp6_type == ICMP6_MEMBERSHIP_QUERY ||
658 icmp6_type == ICMP6_WRUREQUEST ||
659 icmp6_type == ICMP6_FQDN_QUERY ||
660 icmp6_type == ICMP6_NI_QUERY))
667 send_reject6(struct ip_fw_args *args, int code, u_int hlen)
669 if (code == ICMP6_UNREACH_RST && args->f_id.proto == IPPROTO_TCP) {
679 if (args->m->m_len < (hlen+sizeof(struct tcphdr))) {
680 args->m = m_pullup(args->m, hlen+sizeof(struct tcphdr));
685 ip6 = mtod(args->m, struct ip6_hdr *);
686 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
688 if ((tcp->th_flags & TH_RST) != 0) {
695 ti.th.th_seq = ntohl(ti.th.th_seq);
696 ti.th.th_ack = ntohl(ti.th.th_ack);
697 ti.ip6.ip6_nxt = IPPROTO_TCP;
699 if (ti.th.th_flags & TH_ACK) {
705 if (((args->m)->m_flags & M_PKTHDR) != 0) {
706 ack += (args->m)->m_pkthdr.len - hlen
707 - (ti.th.th_off << 2);
708 } else if (ip6->ip6_plen) {
709 ack += ntohs(ip6->ip6_plen) + sizeof(*ip6)
710 - hlen - (ti.th.th_off << 2);
715 if (tcp->th_flags & TH_SYN)
718 flags = TH_RST|TH_ACK;
720 bcopy(&ti, ip6, sizeof(ti));
721 tcp_respond(NULL, ip6, (struct tcphdr *)(ip6 + 1),
722 args->m, ack, seq, flags);
724 } else if (code != ICMP6_UNREACH_RST) { /* Send an ICMPv6 unreach. */
725 icmp6_error(args->m, ICMP6_DST_UNREACH, code, 0);
735 static u_int64_t norule_counter; /* counter for ipfw_log(NULL...) */
737 #define SNPARGS(buf, len) buf + len, sizeof(buf) > len ? sizeof(buf) - len : 0
738 #define SNP(buf) buf, sizeof(buf)
741 * We enter here when we have a rule with O_LOG.
742 * XXX this function alone takes about 2Kbytes of code!
745 ipfw_log(struct ip_fw *f, u_int hlen, struct ip_fw_args *args,
746 struct mbuf *m, struct ifnet *oif, u_short offset)
748 struct ether_header *eh = args->eh;
750 int limit_reached = 0;
751 char action2[40], proto[128], fragment[32];
756 if (f == NULL) { /* bogus pkt */
757 if (verbose_limit != 0 && norule_counter >= verbose_limit)
760 if (norule_counter == verbose_limit)
761 limit_reached = verbose_limit;
763 } else { /* O_LOG is the first action, find the real one */
764 ipfw_insn *cmd = ACTION_PTR(f);
765 ipfw_insn_log *l = (ipfw_insn_log *)cmd;
767 if (l->max_log != 0 && l->log_left == 0)
770 if (l->log_left == 0)
771 limit_reached = l->max_log;
772 cmd += F_LEN(cmd); /* point to first action */
773 if (cmd->opcode == O_ALTQ) {
774 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
776 snprintf(SNPARGS(action2, 0), "Altq %d",
780 if (cmd->opcode == O_PROB)
783 if (cmd->opcode == O_TAG)
787 switch (cmd->opcode) {
793 if (cmd->arg1==ICMP_REJECT_RST)
795 else if (cmd->arg1==ICMP_UNREACH_HOST)
798 snprintf(SNPARGS(action2, 0), "Unreach %d",
803 if (cmd->arg1==ICMP6_UNREACH_RST)
806 snprintf(SNPARGS(action2, 0), "Unreach %d",
817 snprintf(SNPARGS(action2, 0), "Divert %d",
821 snprintf(SNPARGS(action2, 0), "Tee %d",
825 snprintf(SNPARGS(action2, 0), "SkipTo %d",
829 snprintf(SNPARGS(action2, 0), "Pipe %d",
833 snprintf(SNPARGS(action2, 0), "Queue %d",
837 ipfw_insn_sa *sa = (ipfw_insn_sa *)cmd;
840 len = snprintf(SNPARGS(action2, 0), "Forward to %s",
841 inet_ntoa(sa->sa.sin_addr));
843 snprintf(SNPARGS(action2, len), ":%d",
848 snprintf(SNPARGS(action2, 0), "Netgraph %d",
852 snprintf(SNPARGS(action2, 0), "Ngtee %d",
861 if (hlen == 0) { /* non-ip */
862 snprintf(SNPARGS(proto, 0), "MAC");
866 char src[48], dst[48];
867 struct icmphdr *icmp;
870 /* Initialize to make compiler happy. */
871 struct ip *ip = NULL;
873 struct ip6_hdr *ip6 = NULL;
874 struct icmp6_hdr *icmp6;
879 if (args->f_id.addr_type == 6) {
880 snprintf(src, sizeof(src), "[%s]",
881 ip6_sprintf(&args->f_id.src_ip6));
882 snprintf(dst, sizeof(dst), "[%s]",
883 ip6_sprintf(&args->f_id.dst_ip6));
885 ip6 = (struct ip6_hdr *)mtod(m, struct ip6_hdr *);
886 tcp = (struct tcphdr *)(mtod(args->m, char *) + hlen);
887 udp = (struct udphdr *)(mtod(args->m, char *) + hlen);
891 ip = mtod(m, struct ip *);
892 tcp = L3HDR(struct tcphdr, ip);
893 udp = L3HDR(struct udphdr, ip);
895 inet_ntoa_r(ip->ip_src, src);
896 inet_ntoa_r(ip->ip_dst, dst);
899 switch (args->f_id.proto) {
901 len = snprintf(SNPARGS(proto, 0), "TCP %s", src);
903 snprintf(SNPARGS(proto, len), ":%d %s:%d",
904 ntohs(tcp->th_sport),
906 ntohs(tcp->th_dport));
908 snprintf(SNPARGS(proto, len), " %s", dst);
912 len = snprintf(SNPARGS(proto, 0), "UDP %s", src);
914 snprintf(SNPARGS(proto, len), ":%d %s:%d",
915 ntohs(udp->uh_sport),
917 ntohs(udp->uh_dport));
919 snprintf(SNPARGS(proto, len), " %s", dst);
923 icmp = L3HDR(struct icmphdr, ip);
925 len = snprintf(SNPARGS(proto, 0),
927 icmp->icmp_type, icmp->icmp_code);
929 len = snprintf(SNPARGS(proto, 0), "ICMP ");
930 len += snprintf(SNPARGS(proto, len), "%s", src);
931 snprintf(SNPARGS(proto, len), " %s", dst);
935 icmp6 = (struct icmp6_hdr *)(mtod(args->m, char *) + hlen);
937 len = snprintf(SNPARGS(proto, 0),
939 icmp6->icmp6_type, icmp6->icmp6_code);
941 len = snprintf(SNPARGS(proto, 0), "ICMPv6 ");
942 len += snprintf(SNPARGS(proto, len), "%s", src);
943 snprintf(SNPARGS(proto, len), " %s", dst);
947 len = snprintf(SNPARGS(proto, 0), "P:%d %s",
948 args->f_id.proto, src);
949 snprintf(SNPARGS(proto, len), " %s", dst);
954 if (args->f_id.addr_type == 6) {
955 if (offset & (IP6F_OFF_MASK | IP6F_MORE_FRAG))
956 snprintf(SNPARGS(fragment, 0),
957 " (frag %08x:%d@%d%s)",
959 ntohs(ip6->ip6_plen) - hlen,
960 ntohs(offset & IP6F_OFF_MASK) << 3,
961 (offset & IP6F_MORE_FRAG) ? "+" : "");
966 if (eh != NULL) { /* layer 2 packets are as on the wire */
967 ip_off = ntohs(ip->ip_off);
968 ip_len = ntohs(ip->ip_len);
973 if (ip_off & (IP_MF | IP_OFFMASK))
974 snprintf(SNPARGS(fragment, 0),
975 " (frag %d:%d@%d%s)",
976 ntohs(ip->ip_id), ip_len - (ip->ip_hl << 2),
978 (ip_off & IP_MF) ? "+" : "");
981 if (oif || m->m_pkthdr.rcvif)
982 log(LOG_SECURITY | LOG_INFO,
983 "ipfw: %d %s %s %s via %s%s\n",
985 action, proto, oif ? "out" : "in",
986 oif ? oif->if_xname : m->m_pkthdr.rcvif->if_xname,
989 log(LOG_SECURITY | LOG_INFO,
990 "ipfw: %d %s %s [no if info]%s\n",
992 action, proto, fragment);
994 log(LOG_SECURITY | LOG_NOTICE,
995 "ipfw: limit %d reached on entry %d\n",
996 limit_reached, f ? f->rulenum : -1);
1000 * IMPORTANT: the hash function for dynamic rules must be commutative
1001 * in source and destination (ip,port), because rules are bidirectional
1002 * and we want to find both in the same bucket.
1005 hash_packet(struct ipfw_flow_id *id)
1010 if (IS_IP6_FLOW_ID(id))
1011 i = hash_packet6(id);
1014 i = (id->dst_ip) ^ (id->src_ip) ^ (id->dst_port) ^ (id->src_port);
1015 i &= (curr_dyn_buckets - 1);
1020 * unlink a dynamic rule from a chain. prev is a pointer to
1021 * the previous one, q is a pointer to the rule to delete,
1022 * head is a pointer to the head of the queue.
1023 * Modifies q and potentially also head.
1025 #define UNLINK_DYN_RULE(prev, head, q) { \
1026 ipfw_dyn_rule *old_q = q; \
1028 /* remove a refcount to the parent */ \
1029 if (q->dyn_type == O_LIMIT) \
1030 q->parent->count--; \
1031 DEB(printf("ipfw: unlink entry 0x%08x %d -> 0x%08x %d, %d left\n",\
1032 (q->id.src_ip), (q->id.src_port), \
1033 (q->id.dst_ip), (q->id.dst_port), dyn_count-1 ); ) \
1035 prev->next = q = q->next; \
1037 head = q = q->next; \
1039 uma_zfree(ipfw_dyn_rule_zone, old_q); }
1041 #define TIME_LEQ(a,b) ((int)((a)-(b)) <= 0)
1044 * Remove dynamic rules pointing to "rule", or all of them if rule == NULL.
1046 * If keep_me == NULL, rules are deleted even if not expired,
1047 * otherwise only expired rules are removed.
1049 * The value of the second parameter is also used to point to identify
1050 * a rule we absolutely do not want to remove (e.g. because we are
1051 * holding a reference to it -- this is the case with O_LIMIT_PARENT
1052 * rules). The pointer is only used for comparison, so any non-null
1056 remove_dyn_rule(struct ip_fw *rule, ipfw_dyn_rule *keep_me)
1058 static u_int32_t last_remove = 0;
1060 #define FORCE (keep_me == NULL)
1062 ipfw_dyn_rule *prev, *q;
1063 int i, pass = 0, max_pass = 0;
1065 IPFW_DYN_LOCK_ASSERT();
1067 if (ipfw_dyn_v == NULL || dyn_count == 0)
1069 /* do not expire more than once per second, it is useless */
1070 if (!FORCE && last_remove == time_uptime)
1072 last_remove = time_uptime;
1075 * because O_LIMIT refer to parent rules, during the first pass only
1076 * remove child and mark any pending LIMIT_PARENT, and remove
1077 * them in a second pass.
1080 for (i = 0 ; i < curr_dyn_buckets ; i++) {
1081 for (prev=NULL, q = ipfw_dyn_v[i] ; q ; ) {
1083 * Logic can become complex here, so we split tests.
1087 if (rule != NULL && rule != q->rule)
1088 goto next; /* not the one we are looking for */
1089 if (q->dyn_type == O_LIMIT_PARENT) {
1091 * handle parent in the second pass,
1092 * record we need one.
1097 if (FORCE && q->count != 0 ) {
1098 /* XXX should not happen! */
1099 printf("ipfw: OUCH! cannot remove rule,"
1100 " count %d\n", q->count);
1104 !TIME_LEQ( q->expire, time_uptime ))
1107 if (q->dyn_type != O_LIMIT_PARENT || !q->count) {
1108 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1116 if (pass++ < max_pass)
1122 * lookup a dynamic rule.
1124 static ipfw_dyn_rule *
1125 lookup_dyn_rule_locked(struct ipfw_flow_id *pkt, int *match_direction,
1129 * stateful ipfw extensions.
1130 * Lookup into dynamic session queue
1132 #define MATCH_REVERSE 0
1133 #define MATCH_FORWARD 1
1134 #define MATCH_NONE 2
1135 #define MATCH_UNKNOWN 3
1136 int i, dir = MATCH_NONE;
1137 ipfw_dyn_rule *prev, *q=NULL;
1139 IPFW_DYN_LOCK_ASSERT();
1141 if (ipfw_dyn_v == NULL)
1142 goto done; /* not found */
1143 i = hash_packet( pkt );
1144 for (prev=NULL, q = ipfw_dyn_v[i] ; q != NULL ; ) {
1145 if (q->dyn_type == O_LIMIT_PARENT && q->count)
1147 if (TIME_LEQ( q->expire, time_uptime)) { /* expire entry */
1148 UNLINK_DYN_RULE(prev, ipfw_dyn_v[i], q);
1151 if (pkt->proto == q->id.proto &&
1152 q->dyn_type != O_LIMIT_PARENT) {
1153 if (IS_IP6_FLOW_ID(pkt)) {
1154 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1155 &(q->id.src_ip6)) &&
1156 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1157 &(q->id.dst_ip6)) &&
1158 pkt->src_port == q->id.src_port &&
1159 pkt->dst_port == q->id.dst_port ) {
1160 dir = MATCH_FORWARD;
1163 if (IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1164 &(q->id.dst_ip6)) &&
1165 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1166 &(q->id.src_ip6)) &&
1167 pkt->src_port == q->id.dst_port &&
1168 pkt->dst_port == q->id.src_port ) {
1169 dir = MATCH_REVERSE;
1173 if (pkt->src_ip == q->id.src_ip &&
1174 pkt->dst_ip == q->id.dst_ip &&
1175 pkt->src_port == q->id.src_port &&
1176 pkt->dst_port == q->id.dst_port ) {
1177 dir = MATCH_FORWARD;
1180 if (pkt->src_ip == q->id.dst_ip &&
1181 pkt->dst_ip == q->id.src_ip &&
1182 pkt->src_port == q->id.dst_port &&
1183 pkt->dst_port == q->id.src_port ) {
1184 dir = MATCH_REVERSE;
1194 goto done; /* q = NULL, not found */
1196 if ( prev != NULL) { /* found and not in front */
1197 prev->next = q->next;
1198 q->next = ipfw_dyn_v[i];
1201 if (pkt->proto == IPPROTO_TCP) { /* update state according to flags */
1202 u_char flags = pkt->flags & (TH_FIN|TH_SYN|TH_RST);
1204 #define BOTH_SYN (TH_SYN | (TH_SYN << 8))
1205 #define BOTH_FIN (TH_FIN | (TH_FIN << 8))
1206 q->state |= (dir == MATCH_FORWARD ) ? flags : (flags << 8);
1208 case TH_SYN: /* opening */
1209 q->expire = time_uptime + dyn_syn_lifetime;
1212 case BOTH_SYN: /* move to established */
1213 case BOTH_SYN | TH_FIN : /* one side tries to close */
1214 case BOTH_SYN | (TH_FIN << 8) :
1216 #define _SEQ_GE(a,b) ((int)(a) - (int)(b) >= 0)
1217 u_int32_t ack = ntohl(tcp->th_ack);
1218 if (dir == MATCH_FORWARD) {
1219 if (q->ack_fwd == 0 || _SEQ_GE(ack, q->ack_fwd))
1221 else { /* ignore out-of-sequence */
1225 if (q->ack_rev == 0 || _SEQ_GE(ack, q->ack_rev))
1227 else { /* ignore out-of-sequence */
1232 q->expire = time_uptime + dyn_ack_lifetime;
1235 case BOTH_SYN | BOTH_FIN: /* both sides closed */
1236 if (dyn_fin_lifetime >= dyn_keepalive_period)
1237 dyn_fin_lifetime = dyn_keepalive_period - 1;
1238 q->expire = time_uptime + dyn_fin_lifetime;
1244 * reset or some invalid combination, but can also
1245 * occur if we use keep-state the wrong way.
1247 if ( (q->state & ((TH_RST << 8)|TH_RST)) == 0)
1248 printf("invalid state: 0x%x\n", q->state);
1250 if (dyn_rst_lifetime >= dyn_keepalive_period)
1251 dyn_rst_lifetime = dyn_keepalive_period - 1;
1252 q->expire = time_uptime + dyn_rst_lifetime;
1255 } else if (pkt->proto == IPPROTO_UDP) {
1256 q->expire = time_uptime + dyn_udp_lifetime;
1258 /* other protocols */
1259 q->expire = time_uptime + dyn_short_lifetime;
1262 if (match_direction)
1263 *match_direction = dir;
1267 static ipfw_dyn_rule *
1268 lookup_dyn_rule(struct ipfw_flow_id *pkt, int *match_direction,
1274 q = lookup_dyn_rule_locked(pkt, match_direction, tcp);
1277 /* NB: return table locked when q is not NULL */
1282 realloc_dynamic_table(void)
1284 IPFW_DYN_LOCK_ASSERT();
1287 * Try reallocation, make sure we have a power of 2 and do
1288 * not allow more than 64k entries. In case of overflow,
1292 if (dyn_buckets > 65536)
1294 if ((dyn_buckets & (dyn_buckets-1)) != 0) { /* not a power of 2 */
1295 dyn_buckets = curr_dyn_buckets; /* reset */
1298 curr_dyn_buckets = dyn_buckets;
1299 if (ipfw_dyn_v != NULL)
1300 free(ipfw_dyn_v, M_IPFW);
1302 ipfw_dyn_v = malloc(curr_dyn_buckets * sizeof(ipfw_dyn_rule *),
1303 M_IPFW, M_NOWAIT | M_ZERO);
1304 if (ipfw_dyn_v != NULL || curr_dyn_buckets <= 2)
1306 curr_dyn_buckets /= 2;
1311 * Install state of type 'type' for a dynamic session.
1312 * The hash table contains two type of rules:
1313 * - regular rules (O_KEEP_STATE)
1314 * - rules for sessions with limited number of sess per user
1315 * (O_LIMIT). When they are created, the parent is
1316 * increased by 1, and decreased on delete. In this case,
1317 * the third parameter is the parent rule and not the chain.
1318 * - "parent" rules for the above (O_LIMIT_PARENT).
1320 static ipfw_dyn_rule *
1321 add_dyn_rule(struct ipfw_flow_id *id, u_int8_t dyn_type, struct ip_fw *rule)
1326 IPFW_DYN_LOCK_ASSERT();
1328 if (ipfw_dyn_v == NULL ||
1329 (dyn_count == 0 && dyn_buckets != curr_dyn_buckets)) {
1330 realloc_dynamic_table();
1331 if (ipfw_dyn_v == NULL)
1332 return NULL; /* failed ! */
1334 i = hash_packet(id);
1336 r = uma_zalloc(ipfw_dyn_rule_zone, M_NOWAIT | M_ZERO);
1338 printf ("ipfw: sorry cannot allocate state\n");
1342 /* increase refcount on parent, and set pointer */
1343 if (dyn_type == O_LIMIT) {
1344 ipfw_dyn_rule *parent = (ipfw_dyn_rule *)rule;
1345 if ( parent->dyn_type != O_LIMIT_PARENT)
1346 panic("invalid parent");
1349 rule = parent->rule;
1353 r->expire = time_uptime + dyn_syn_lifetime;
1355 r->dyn_type = dyn_type;
1356 r->pcnt = r->bcnt = 0;
1360 r->next = ipfw_dyn_v[i];
1363 DEB(printf("ipfw: add dyn entry ty %d 0x%08x %d -> 0x%08x %d, total %d\n",
1365 (r->id.src_ip), (r->id.src_port),
1366 (r->id.dst_ip), (r->id.dst_port),
1372 * lookup dynamic parent rule using pkt and rule as search keys.
1373 * If the lookup fails, then install one.
1375 static ipfw_dyn_rule *
1376 lookup_dyn_parent(struct ipfw_flow_id *pkt, struct ip_fw *rule)
1381 IPFW_DYN_LOCK_ASSERT();
1384 int is_v6 = IS_IP6_FLOW_ID(pkt);
1385 i = hash_packet( pkt );
1386 for (q = ipfw_dyn_v[i] ; q != NULL ; q=q->next)
1387 if (q->dyn_type == O_LIMIT_PARENT &&
1389 pkt->proto == q->id.proto &&
1390 pkt->src_port == q->id.src_port &&
1391 pkt->dst_port == q->id.dst_port &&
1394 IN6_ARE_ADDR_EQUAL(&(pkt->src_ip6),
1395 &(q->id.src_ip6)) &&
1396 IN6_ARE_ADDR_EQUAL(&(pkt->dst_ip6),
1397 &(q->id.dst_ip6))) ||
1399 pkt->src_ip == q->id.src_ip &&
1400 pkt->dst_ip == q->id.dst_ip)
1403 q->expire = time_uptime + dyn_short_lifetime;
1404 DEB(printf("ipfw: lookup_dyn_parent found 0x%p\n",q);)
1408 return add_dyn_rule(pkt, O_LIMIT_PARENT, rule);
1412 * Install dynamic state for rule type cmd->o.opcode
1414 * Returns 1 (failure) if state is not installed because of errors or because
1415 * session limitations are enforced.
1418 install_state(struct ip_fw *rule, ipfw_insn_limit *cmd,
1419 struct ip_fw_args *args, uint32_t tablearg)
1421 static int last_log;
1426 printf("ipfw: %s: type %d 0x%08x %u -> 0x%08x %u\n",
1427 __func__, cmd->o.opcode,
1428 (args->f_id.src_ip), (args->f_id.src_port),
1429 (args->f_id.dst_ip), (args->f_id.dst_port));
1434 q = lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1436 if (q != NULL) { /* should never occur */
1437 if (last_log != time_uptime) {
1438 last_log = time_uptime;
1439 printf("ipfw: %s: entry already present, done\n",
1446 if (dyn_count >= dyn_max)
1447 /* Run out of slots, try to remove any expired rule. */
1448 remove_dyn_rule(NULL, (ipfw_dyn_rule *)1);
1450 if (dyn_count >= dyn_max) {
1451 if (last_log != time_uptime) {
1452 last_log = time_uptime;
1453 printf("ipfw: %s: Too many dynamic rules\n", __func__);
1456 return (1); /* cannot install, notify caller */
1459 switch (cmd->o.opcode) {
1460 case O_KEEP_STATE: /* bidir rule */
1461 add_dyn_rule(&args->f_id, O_KEEP_STATE, rule);
1464 case O_LIMIT: { /* limit number of sessions */
1465 struct ipfw_flow_id id;
1466 ipfw_dyn_rule *parent;
1467 uint32_t conn_limit;
1468 uint16_t limit_mask = cmd->limit_mask;
1470 conn_limit = (cmd->conn_limit == IP_FW_TABLEARG) ?
1471 tablearg : cmd->conn_limit;
1474 if (cmd->conn_limit == IP_FW_TABLEARG)
1475 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u "
1476 "(tablearg)\n", __func__, conn_limit);
1478 printf("ipfw: %s: O_LIMIT rule, conn_limit: %u\n",
1479 __func__, conn_limit);
1482 id.dst_ip = id.src_ip = id.dst_port = id.src_port = 0;
1483 id.proto = args->f_id.proto;
1484 id.addr_type = args->f_id.addr_type;
1486 if (IS_IP6_FLOW_ID (&(args->f_id))) {
1487 if (limit_mask & DYN_SRC_ADDR)
1488 id.src_ip6 = args->f_id.src_ip6;
1489 if (limit_mask & DYN_DST_ADDR)
1490 id.dst_ip6 = args->f_id.dst_ip6;
1492 if (limit_mask & DYN_SRC_ADDR)
1493 id.src_ip = args->f_id.src_ip;
1494 if (limit_mask & DYN_DST_ADDR)
1495 id.dst_ip = args->f_id.dst_ip;
1497 if (limit_mask & DYN_SRC_PORT)
1498 id.src_port = args->f_id.src_port;
1499 if (limit_mask & DYN_DST_PORT)
1500 id.dst_port = args->f_id.dst_port;
1501 if ((parent = lookup_dyn_parent(&id, rule)) == NULL) {
1502 printf("ipfw: %s: add parent failed\n", __func__);
1507 if (parent->count >= conn_limit) {
1508 /* See if we can remove some expired rule. */
1509 remove_dyn_rule(rule, parent);
1510 if (parent->count >= conn_limit) {
1511 if (fw_verbose && last_log != time_uptime) {
1512 last_log = time_uptime;
1513 log(LOG_SECURITY | LOG_DEBUG,
1514 "drop session, too many entries\n");
1520 add_dyn_rule(&args->f_id, O_LIMIT, (struct ip_fw *)parent);
1524 printf("ipfw: %s: unknown dynamic rule type %u\n",
1525 __func__, cmd->o.opcode);
1530 /* XXX just set lifetime */
1531 lookup_dyn_rule_locked(&args->f_id, NULL, NULL);
1538 * Generate a TCP packet, containing either a RST or a keepalive.
1539 * When flags & TH_RST, we are sending a RST packet, because of a
1540 * "reset" action matched the packet.
1541 * Otherwise we are sending a keepalive, and flags & TH_
1543 static struct mbuf *
1544 send_pkt(struct ipfw_flow_id *id, u_int32_t seq, u_int32_t ack, int flags)
1550 MGETHDR(m, M_DONTWAIT, MT_DATA);
1553 m->m_pkthdr.rcvif = (struct ifnet *)0;
1554 m->m_pkthdr.len = m->m_len = sizeof(struct ip) + sizeof(struct tcphdr);
1555 m->m_data += max_linkhdr;
1557 ip = mtod(m, struct ip *);
1558 bzero(ip, m->m_len);
1559 tcp = (struct tcphdr *)(ip + 1); /* no IP options */
1560 ip->ip_p = IPPROTO_TCP;
1563 * Assume we are sending a RST (or a keepalive in the reverse
1564 * direction), swap src and destination addresses and ports.
1566 ip->ip_src.s_addr = htonl(id->dst_ip);
1567 ip->ip_dst.s_addr = htonl(id->src_ip);
1568 tcp->th_sport = htons(id->dst_port);
1569 tcp->th_dport = htons(id->src_port);
1570 if (flags & TH_RST) { /* we are sending a RST */
1571 if (flags & TH_ACK) {
1572 tcp->th_seq = htonl(ack);
1573 tcp->th_ack = htonl(0);
1574 tcp->th_flags = TH_RST;
1578 tcp->th_seq = htonl(0);
1579 tcp->th_ack = htonl(seq);
1580 tcp->th_flags = TH_RST | TH_ACK;
1584 * We are sending a keepalive. flags & TH_SYN determines
1585 * the direction, forward if set, reverse if clear.
1586 * NOTE: seq and ack are always assumed to be correct
1587 * as set by the caller. This may be confusing...
1589 if (flags & TH_SYN) {
1591 * we have to rewrite the correct addresses!
1593 ip->ip_dst.s_addr = htonl(id->dst_ip);
1594 ip->ip_src.s_addr = htonl(id->src_ip);
1595 tcp->th_dport = htons(id->dst_port);
1596 tcp->th_sport = htons(id->src_port);
1598 tcp->th_seq = htonl(seq);
1599 tcp->th_ack = htonl(ack);
1600 tcp->th_flags = TH_ACK;
1603 * set ip_len to the payload size so we can compute
1604 * the tcp checksum on the pseudoheader
1605 * XXX check this, could save a couple of words ?
1607 ip->ip_len = htons(sizeof(struct tcphdr));
1608 tcp->th_sum = in_cksum(m, m->m_pkthdr.len);
1610 * now fill fields left out earlier
1612 ip->ip_ttl = ip_defttl;
1613 ip->ip_len = m->m_pkthdr.len;
1614 m->m_flags |= M_SKIP_FIREWALL;
1619 * sends a reject message, consuming the mbuf passed as an argument.
1622 send_reject(struct ip_fw_args *args, int code, int ip_len)
1625 if (code != ICMP_REJECT_RST) { /* Send an ICMP unreach */
1626 /* We need the IP header in host order for icmp_error(). */
1627 if (args->eh != NULL) {
1628 struct ip *ip = mtod(args->m, struct ip *);
1629 ip->ip_len = ntohs(ip->ip_len);
1630 ip->ip_off = ntohs(ip->ip_off);
1632 icmp_error(args->m, ICMP_UNREACH, code, 0L, 0);
1633 } else if (args->f_id.proto == IPPROTO_TCP) {
1634 struct tcphdr *const tcp =
1635 L3HDR(struct tcphdr, mtod(args->m, struct ip *));
1636 if ( (tcp->th_flags & TH_RST) == 0) {
1638 m = send_pkt(&(args->f_id), ntohl(tcp->th_seq),
1640 tcp->th_flags | TH_RST);
1642 ip_output(m, NULL, NULL, 0, NULL, NULL);
1652 * Given an ip_fw *, lookup_next_rule will return a pointer
1653 * to the next rule, which can be either the jump
1654 * target (for skipto instructions) or the next one in the list (in
1655 * all other cases including a missing jump target).
1656 * The result is also written in the "next_rule" field of the rule.
1657 * Backward jumps are not allowed, so start looking from the next
1660 * This never returns NULL -- in case we do not have an exact match,
1661 * the next rule is returned. When the ruleset is changed,
1662 * pointers are flushed so we are always correct.
1665 static struct ip_fw *
1666 lookup_next_rule(struct ip_fw *me)
1668 struct ip_fw *rule = NULL;
1671 /* look for action, in case it is a skipto */
1672 cmd = ACTION_PTR(me);
1673 if (cmd->opcode == O_LOG)
1675 if (cmd->opcode == O_ALTQ)
1677 if (cmd->opcode == O_TAG)
1679 if ( cmd->opcode == O_SKIPTO )
1680 for (rule = me->next; rule ; rule = rule->next)
1681 if (rule->rulenum >= cmd->arg1)
1683 if (rule == NULL) /* failure or not a skipto */
1685 me->next_rule = rule;
1690 add_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1691 uint8_t mlen, uint32_t value)
1693 struct radix_node_head *rnh;
1694 struct table_entry *ent;
1696 if (tbl >= IPFW_TABLES_MAX)
1698 rnh = ch->tables[tbl];
1699 ent = malloc(sizeof(*ent), M_IPFW_TBL, M_NOWAIT | M_ZERO);
1703 ent->addr.sin_len = ent->mask.sin_len = 8;
1704 ent->mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1705 ent->addr.sin_addr.s_addr = addr & ent->mask.sin_addr.s_addr;
1706 IPFW_WLOCK(&layer3_chain);
1707 if (rnh->rnh_addaddr(&ent->addr, &ent->mask, rnh, (void *)ent) ==
1709 IPFW_WUNLOCK(&layer3_chain);
1710 free(ent, M_IPFW_TBL);
1713 IPFW_WUNLOCK(&layer3_chain);
1718 del_table_entry(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1721 struct radix_node_head *rnh;
1722 struct table_entry *ent;
1723 struct sockaddr_in sa, mask;
1725 if (tbl >= IPFW_TABLES_MAX)
1727 rnh = ch->tables[tbl];
1728 sa.sin_len = mask.sin_len = 8;
1729 mask.sin_addr.s_addr = htonl(mlen ? ~((1 << (32 - mlen)) - 1) : 0);
1730 sa.sin_addr.s_addr = addr & mask.sin_addr.s_addr;
1732 ent = (struct table_entry *)rnh->rnh_deladdr(&sa, &mask, rnh);
1738 free(ent, M_IPFW_TBL);
1743 flush_table_entry(struct radix_node *rn, void *arg)
1745 struct radix_node_head * const rnh = arg;
1746 struct table_entry *ent;
1748 ent = (struct table_entry *)
1749 rnh->rnh_deladdr(rn->rn_key, rn->rn_mask, rnh);
1751 free(ent, M_IPFW_TBL);
1756 flush_table(struct ip_fw_chain *ch, uint16_t tbl)
1758 struct radix_node_head *rnh;
1760 IPFW_WLOCK_ASSERT(ch);
1762 if (tbl >= IPFW_TABLES_MAX)
1764 rnh = ch->tables[tbl];
1765 KASSERT(rnh != NULL, ("NULL IPFW table"));
1766 rnh->rnh_walktree(rnh, flush_table_entry, rnh);
1771 flush_tables(struct ip_fw_chain *ch)
1775 IPFW_WLOCK_ASSERT(ch);
1777 for (tbl = 0; tbl < IPFW_TABLES_MAX; tbl++)
1778 flush_table(ch, tbl);
1782 init_tables(struct ip_fw_chain *ch)
1787 for (i = 0; i < IPFW_TABLES_MAX; i++) {
1788 if (!rn_inithead((void **)&ch->tables[i], 32)) {
1789 for (j = 0; j < i; j++) {
1790 (void) flush_table(ch, j);
1799 lookup_table(struct ip_fw_chain *ch, uint16_t tbl, in_addr_t addr,
1802 struct radix_node_head *rnh;
1803 struct table_entry *ent;
1804 struct sockaddr_in sa;
1806 if (tbl >= IPFW_TABLES_MAX)
1808 rnh = ch->tables[tbl];
1810 sa.sin_addr.s_addr = addr;
1811 ent = (struct table_entry *)(rnh->rnh_lookup(&sa, NULL, rnh));
1820 count_table_entry(struct radix_node *rn, void *arg)
1822 u_int32_t * const cnt = arg;
1829 count_table(struct ip_fw_chain *ch, uint32_t tbl, uint32_t *cnt)
1831 struct radix_node_head *rnh;
1833 if (tbl >= IPFW_TABLES_MAX)
1835 rnh = ch->tables[tbl];
1837 rnh->rnh_walktree(rnh, count_table_entry, cnt);
1842 dump_table_entry(struct radix_node *rn, void *arg)
1844 struct table_entry * const n = (struct table_entry *)rn;
1845 ipfw_table * const tbl = arg;
1846 ipfw_table_entry *ent;
1848 if (tbl->cnt == tbl->size)
1850 ent = &tbl->ent[tbl->cnt];
1851 ent->tbl = tbl->tbl;
1852 if (in_nullhost(n->mask.sin_addr))
1855 ent->masklen = 33 - ffs(ntohl(n->mask.sin_addr.s_addr));
1856 ent->addr = n->addr.sin_addr.s_addr;
1857 ent->value = n->value;
1863 dump_table(struct ip_fw_chain *ch, ipfw_table *tbl)
1865 struct radix_node_head *rnh;
1867 if (tbl->tbl >= IPFW_TABLES_MAX)
1869 rnh = ch->tables[tbl->tbl];
1871 rnh->rnh_walktree(rnh, dump_table_entry, tbl);
1876 fill_ugid_cache(struct inpcb *inp, struct ip_fw_ugid *ugp)
1880 if (inp->inp_socket != NULL) {
1881 cr = inp->inp_socket->so_cred;
1882 ugp->fw_prid = jailed(cr) ?
1883 cr->cr_prison->pr_id : -1;
1884 ugp->fw_uid = cr->cr_uid;
1885 ugp->fw_ngroups = cr->cr_ngroups;
1886 bcopy(cr->cr_groups, ugp->fw_groups,
1887 sizeof(ugp->fw_groups));
1892 check_uidgid(ipfw_insn_u32 *insn,
1893 int proto, struct ifnet *oif,
1894 struct in_addr dst_ip, u_int16_t dst_port,
1895 struct in_addr src_ip, u_int16_t src_port,
1896 struct ip_fw_ugid *ugp, int *lookup, struct inpcb *inp)
1898 struct inpcbinfo *pi;
1905 * Check to see if the UDP or TCP stack supplied us with
1906 * the PCB. If so, rather then holding a lock and looking
1907 * up the PCB, we can use the one that was supplied.
1909 if (inp && *lookup == 0) {
1910 INP_LOCK_ASSERT(inp);
1911 if (inp->inp_socket != NULL) {
1912 fill_ugid_cache(inp, ugp);
1917 * If we have already been here and the packet has no
1918 * PCB entry associated with it, then we can safely
1919 * assume that this is a no match.
1923 if (proto == IPPROTO_TCP) {
1926 } else if (proto == IPPROTO_UDP) {
1927 wildcard = INPLOOKUP_WILDCARD;
1935 in_pcblookup_hash(pi,
1936 dst_ip, htons(dst_port),
1937 src_ip, htons(src_port),
1939 in_pcblookup_hash(pi,
1940 src_ip, htons(src_port),
1941 dst_ip, htons(dst_port),
1945 if (pcb->inp_socket != NULL) {
1946 fill_ugid_cache(pcb, ugp);
1951 INP_INFO_RUNLOCK(pi);
1954 * If the lookup did not yield any results, there
1955 * is no sense in coming back and trying again. So
1956 * we can set lookup to -1 and ensure that we wont
1957 * bother the pcb system again.
1963 if (insn->o.opcode == O_UID)
1964 match = (ugp->fw_uid == (uid_t)insn->d[0]);
1965 else if (insn->o.opcode == O_GID) {
1966 for (gp = ugp->fw_groups;
1967 gp < &ugp->fw_groups[ugp->fw_ngroups]; gp++)
1968 if (*gp == (gid_t)insn->d[0]) {
1972 } else if (insn->o.opcode == O_JAIL)
1973 match = (ugp->fw_prid == (int)insn->d[0]);
1978 * The main check routine for the firewall.
1980 * All arguments are in args so we can modify them and return them
1981 * back to the caller.
1985 * args->m (in/out) The packet; we set to NULL when/if we nuke it.
1986 * Starts with the IP header.
1987 * args->eh (in) Mac header if present, or NULL for layer3 packet.
1988 * args->oif Outgoing interface, or NULL if packet is incoming.
1989 * The incoming interface is in the mbuf. (in)
1990 * args->divert_rule (in/out)
1991 * Skip up to the first rule past this rule number;
1992 * upon return, non-zero port number for divert or tee.
1994 * args->rule Pointer to the last matching rule (in/out)
1995 * args->next_hop Socket we are forwarding to (out).
1996 * args->f_id Addresses grabbed from the packet (out)
1997 * args->cookie a cookie depending on rule action
2001 * IP_FW_PASS the packet must be accepted
2002 * IP_FW_DENY the packet must be dropped
2003 * IP_FW_DIVERT divert packet, port in m_tag
2004 * IP_FW_TEE tee packet, port in m_tag
2005 * IP_FW_DUMMYNET to dummynet, pipe in args->cookie
2006 * IP_FW_NETGRAPH into netgraph, cookie args->cookie
2011 ipfw_chk(struct ip_fw_args *args)
2014 * Local variables hold state during the processing of a packet.
2016 * IMPORTANT NOTE: to speed up the processing of rules, there
2017 * are some assumption on the values of the variables, which
2018 * are documented here. Should you change them, please check
2019 * the implementation of the various instructions to make sure
2020 * that they still work.
2022 * args->eh The MAC header. It is non-null for a layer2
2023 * packet, it is NULL for a layer-3 packet.
2025 * m | args->m Pointer to the mbuf, as received from the caller.
2026 * It may change if ipfw_chk() does an m_pullup, or if it
2027 * consumes the packet because it calls send_reject().
2028 * XXX This has to change, so that ipfw_chk() never modifies
2029 * or consumes the buffer.
2030 * ip is simply an alias of the value of m, and it is kept
2031 * in sync with it (the packet is supposed to start with
2034 struct mbuf *m = args->m;
2035 struct ip *ip = mtod(m, struct ip *);
2038 * For rules which contain uid/gid or jail constraints, cache
2039 * a copy of the users credentials after the pcb lookup has been
2040 * executed. This will speed up the processing of rules with
2041 * these types of constraints, as well as decrease contention
2042 * on pcb related locks.
2044 struct ip_fw_ugid fw_ugid_cache;
2045 int ugid_lookup = 0;
2048 * divinput_flags If non-zero, set to the IP_FW_DIVERT_*_FLAG
2049 * associated with a packet input on a divert socket. This
2050 * will allow to distinguish traffic and its direction when
2051 * it originates from a divert socket.
2053 u_int divinput_flags = 0;
2056 * oif | args->oif If NULL, ipfw_chk has been called on the
2057 * inbound path (ether_input, ip_input).
2058 * If non-NULL, ipfw_chk has been called on the outbound path
2059 * (ether_output, ip_output).
2061 struct ifnet *oif = args->oif;
2063 struct ip_fw *f = NULL; /* matching rule */
2067 * hlen The length of the IP header.
2069 u_int hlen = 0; /* hlen >0 means we have an IP pkt */
2072 * offset The offset of a fragment. offset != 0 means that
2073 * we have a fragment at this offset of an IPv4 packet.
2074 * offset == 0 means that (if this is an IPv4 packet)
2075 * this is the first or only fragment.
2076 * For IPv6 offset == 0 means there is no Fragment Header.
2077 * If offset != 0 for IPv6 always use correct mask to
2078 * get the correct offset because we add IP6F_MORE_FRAG
2079 * to be able to dectect the first fragment which would
2080 * otherwise have offset = 0.
2085 * Local copies of addresses. They are only valid if we have
2088 * proto The protocol. Set to 0 for non-ip packets,
2089 * or to the protocol read from the packet otherwise.
2090 * proto != 0 means that we have an IPv4 packet.
2092 * src_port, dst_port port numbers, in HOST format. Only
2093 * valid for TCP and UDP packets.
2095 * src_ip, dst_ip ip addresses, in NETWORK format.
2096 * Only valid for IPv4 packets.
2099 u_int16_t src_port = 0, dst_port = 0; /* NOTE: host format */
2100 struct in_addr src_ip, dst_ip; /* NOTE: network format */
2105 * dyn_dir = MATCH_UNKNOWN when rules unchecked,
2106 * MATCH_NONE when checked and not matched (q = NULL),
2107 * MATCH_FORWARD or MATCH_REVERSE otherwise (q != NULL)
2109 int dyn_dir = MATCH_UNKNOWN;
2110 ipfw_dyn_rule *q = NULL;
2111 struct ip_fw_chain *chain = &layer3_chain;
2115 * We store in ulp a pointer to the upper layer protocol header.
2116 * In the ipv4 case this is easy to determine from the header,
2117 * but for ipv6 we might have some additional headers in the middle.
2118 * ulp is NULL if not found.
2120 void *ulp = NULL; /* upper layer protocol pointer. */
2121 /* XXX ipv6 variables */
2123 u_int16_t ext_hd = 0; /* bits vector for extension header filtering */
2124 /* end of ipv6 variables */
2127 if (m->m_flags & M_SKIP_FIREWALL)
2128 return (IP_FW_PASS); /* accept */
2130 pktlen = m->m_pkthdr.len;
2131 proto = args->f_id.proto = 0; /* mark f_id invalid */
2132 /* XXX 0 is a valid proto: IP/IPv6 Hop-by-Hop Option */
2135 * PULLUP_TO(len, p, T) makes sure that len + sizeof(T) is contiguous,
2136 * then it sets p to point at the offset "len" in the mbuf. WARNING: the
2137 * pointer might become stale after other pullups (but we never use it
2140 #define PULLUP_TO(len, p, T) \
2142 int x = (len) + sizeof(T); \
2143 if ((m)->m_len < x) { \
2144 args->m = m = m_pullup(m, x); \
2146 goto pullup_failed; \
2148 p = (mtod(m, char *) + (len)); \
2151 /* Identify IP packets and fill up variables. */
2152 if (pktlen >= sizeof(struct ip6_hdr) &&
2153 (args->eh == NULL || ntohs(args->eh->ether_type)==ETHERTYPE_IPV6) &&
2154 mtod(m, struct ip *)->ip_v == 6) {
2156 args->f_id.addr_type = 6;
2157 hlen = sizeof(struct ip6_hdr);
2158 proto = mtod(m, struct ip6_hdr *)->ip6_nxt;
2160 /* Search extension headers to find upper layer protocols */
2161 while (ulp == NULL) {
2163 case IPPROTO_ICMPV6:
2164 PULLUP_TO(hlen, ulp, struct icmp6_hdr);
2165 args->f_id.flags = ICMP6(ulp)->icmp6_type;
2169 PULLUP_TO(hlen, ulp, struct tcphdr);
2170 dst_port = TCP(ulp)->th_dport;
2171 src_port = TCP(ulp)->th_sport;
2172 args->f_id.flags = TCP(ulp)->th_flags;
2176 PULLUP_TO(hlen, ulp, struct udphdr);
2177 dst_port = UDP(ulp)->uh_dport;
2178 src_port = UDP(ulp)->uh_sport;
2181 case IPPROTO_HOPOPTS: /* RFC 2460 */
2182 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2183 ext_hd |= EXT_HOPOPTS;
2184 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2185 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2189 case IPPROTO_ROUTING: /* RFC 2460 */
2190 PULLUP_TO(hlen, ulp, struct ip6_rthdr);
2191 switch (((struct ip6_rthdr *)ulp)->ip6r_type) {
2195 printf("IPFW2: IPV6 - Unknown Routing "
2196 "Header type(%d)\n",
2197 ((struct ip6_rthdr *)ulp)->ip6r_type);
2198 if (fw_deny_unknown_exthdrs)
2199 return (IP_FW_DENY);
2202 ext_hd |= EXT_ROUTING;
2203 hlen += (((struct ip6_rthdr *)ulp)->ip6r_len + 1) << 3;
2204 proto = ((struct ip6_rthdr *)ulp)->ip6r_nxt;
2208 case IPPROTO_FRAGMENT: /* RFC 2460 */
2209 PULLUP_TO(hlen, ulp, struct ip6_frag);
2210 ext_hd |= EXT_FRAGMENT;
2211 hlen += sizeof (struct ip6_frag);
2212 proto = ((struct ip6_frag *)ulp)->ip6f_nxt;
2213 offset = ((struct ip6_frag *)ulp)->ip6f_offlg &
2215 /* Add IP6F_MORE_FRAG for offset of first
2216 * fragment to be != 0. */
2217 offset |= ((struct ip6_frag *)ulp)->ip6f_offlg &
2220 printf("IPFW2: IPV6 - Invalid Fragment "
2222 if (fw_deny_unknown_exthdrs)
2223 return (IP_FW_DENY);
2226 args->f_id.frag_id6 =
2227 ntohl(((struct ip6_frag *)ulp)->ip6f_ident);
2231 case IPPROTO_DSTOPTS: /* RFC 2460 */
2232 PULLUP_TO(hlen, ulp, struct ip6_hbh);
2233 ext_hd |= EXT_DSTOPTS;
2234 hlen += (((struct ip6_hbh *)ulp)->ip6h_len + 1) << 3;
2235 proto = ((struct ip6_hbh *)ulp)->ip6h_nxt;
2239 case IPPROTO_AH: /* RFC 2402 */
2240 PULLUP_TO(hlen, ulp, struct ip6_ext);
2242 hlen += (((struct ip6_ext *)ulp)->ip6e_len + 2) << 2;
2243 proto = ((struct ip6_ext *)ulp)->ip6e_nxt;
2247 case IPPROTO_ESP: /* RFC 2406 */
2248 PULLUP_TO(hlen, ulp, uint32_t); /* SPI, Seq# */
2249 /* Anything past Seq# is variable length and
2250 * data past this ext. header is encrypted. */
2254 case IPPROTO_NONE: /* RFC 2460 */
2255 PULLUP_TO(hlen, ulp, struct ip6_ext);
2256 /* Packet ends here. if ip6e_len!=0 octets
2257 * must be ignored. */
2260 case IPPROTO_OSPFIGP:
2261 /* XXX OSPF header check? */
2262 PULLUP_TO(hlen, ulp, struct ip6_ext);
2265 case IPPROTO_IPV6: /* RFC 2893 */
2266 PULLUP_TO(hlen, ulp, struct ip6_hdr);
2269 case IPPROTO_IPV4: /* RFC 2893 */
2270 PULLUP_TO(hlen, ulp, struct ip);
2274 printf("IPFW2: IPV6 - Unknown Extension "
2275 "Header(%d), ext_hd=%x\n", proto, ext_hd);
2276 if (fw_deny_unknown_exthdrs)
2277 return (IP_FW_DENY);
2278 PULLUP_TO(hlen, ulp, struct ip6_ext);
2282 args->f_id.src_ip6 = mtod(m,struct ip6_hdr *)->ip6_src;
2283 args->f_id.dst_ip6 = mtod(m,struct ip6_hdr *)->ip6_dst;
2284 args->f_id.src_ip = 0;
2285 args->f_id.dst_ip = 0;
2286 args->f_id.flow_id6 = ntohl(mtod(m, struct ip6_hdr *)->ip6_flow);
2287 } else if (pktlen >= sizeof(struct ip) &&
2288 (args->eh == NULL || ntohs(args->eh->ether_type) == ETHERTYPE_IP) &&
2289 mtod(m, struct ip *)->ip_v == 4) {
2291 ip = mtod(m, struct ip *);
2292 hlen = ip->ip_hl << 2;
2293 args->f_id.addr_type = 4;
2296 * Collect parameters into local variables for faster matching.
2299 src_ip = ip->ip_src;
2300 dst_ip = ip->ip_dst;
2301 if (args->eh != NULL) { /* layer 2 packets are as on the wire */
2302 offset = ntohs(ip->ip_off) & IP_OFFMASK;
2303 ip_len = ntohs(ip->ip_len);
2305 offset = ip->ip_off & IP_OFFMASK;
2306 ip_len = ip->ip_len;
2308 pktlen = ip_len < pktlen ? ip_len : pktlen;
2313 PULLUP_TO(hlen, ulp, struct tcphdr);
2314 dst_port = TCP(ulp)->th_dport;
2315 src_port = TCP(ulp)->th_sport;
2316 args->f_id.flags = TCP(ulp)->th_flags;
2320 PULLUP_TO(hlen, ulp, struct udphdr);
2321 dst_port = UDP(ulp)->uh_dport;
2322 src_port = UDP(ulp)->uh_sport;
2326 PULLUP_TO(hlen, ulp, struct icmphdr);
2327 args->f_id.flags = ICMP(ulp)->icmp_type;
2335 args->f_id.src_ip = ntohl(src_ip.s_addr);
2336 args->f_id.dst_ip = ntohl(dst_ip.s_addr);
2339 if (proto) { /* we may have port numbers, store them */
2340 args->f_id.proto = proto;
2341 args->f_id.src_port = src_port = ntohs(src_port);
2342 args->f_id.dst_port = dst_port = ntohs(dst_port);
2346 mtag = m_tag_find(m, PACKET_TAG_DIVERT, NULL);
2349 * Packet has already been tagged. Look for the next rule
2350 * to restart processing.
2352 * If fw_one_pass != 0 then just accept it.
2353 * XXX should not happen here, but optimized out in
2357 IPFW_RUNLOCK(chain);
2358 return (IP_FW_PASS);
2361 f = args->rule->next_rule;
2363 f = lookup_next_rule(args->rule);
2366 * Find the starting rule. It can be either the first
2367 * one, or the one after divert_rule if asked so.
2369 int skipto = mtag ? divert_cookie(mtag) : 0;
2372 if (args->eh == NULL && skipto != 0) {
2373 if (skipto >= IPFW_DEFAULT_RULE) {
2374 IPFW_RUNLOCK(chain);
2375 return (IP_FW_DENY); /* invalid */
2377 while (f && f->rulenum <= skipto)
2379 if (f == NULL) { /* drop packet */
2380 IPFW_RUNLOCK(chain);
2381 return (IP_FW_DENY);
2385 /* reset divert rule to avoid confusion later */
2387 divinput_flags = divert_info(mtag) &
2388 (IP_FW_DIVERT_OUTPUT_FLAG | IP_FW_DIVERT_LOOPBACK_FLAG);
2389 m_tag_delete(m, mtag);
2393 * Now scan the rules, and parse microinstructions for each rule.
2395 for (; f; f = f->next) {
2397 uint32_t tablearg = 0;
2398 int l, cmdlen, skip_or; /* skip rest of OR block */
2401 if (set_disable & (1 << f->set) )
2405 for (l = f->cmd_len, cmd = f->cmd ; l > 0 ;
2406 l -= cmdlen, cmd += cmdlen) {
2410 * check_body is a jump target used when we find a
2411 * CHECK_STATE, and need to jump to the body of
2416 cmdlen = F_LEN(cmd);
2418 * An OR block (insn_1 || .. || insn_n) has the
2419 * F_OR bit set in all but the last instruction.
2420 * The first match will set "skip_or", and cause
2421 * the following instructions to be skipped until
2422 * past the one with the F_OR bit clear.
2424 if (skip_or) { /* skip this instruction */
2425 if ((cmd->len & F_OR) == 0)
2426 skip_or = 0; /* next one is good */
2429 match = 0; /* set to 1 if we succeed */
2431 switch (cmd->opcode) {
2433 * The first set of opcodes compares the packet's
2434 * fields with some pattern, setting 'match' if a
2435 * match is found. At the end of the loop there is
2436 * logic to deal with F_NOT and F_OR flags associated
2444 printf("ipfw: opcode %d unimplemented\n",
2452 * We only check offset == 0 && proto != 0,
2453 * as this ensures that we have a
2454 * packet with the ports info.
2458 if (is_ipv6) /* XXX to be fixed later */
2460 if (proto == IPPROTO_TCP ||
2461 proto == IPPROTO_UDP)
2462 match = check_uidgid(
2463 (ipfw_insn_u32 *)cmd,
2466 src_ip, src_port, &fw_ugid_cache,
2467 &ugid_lookup, args->inp);
2471 match = iface_match(m->m_pkthdr.rcvif,
2472 (ipfw_insn_if *)cmd);
2476 match = iface_match(oif, (ipfw_insn_if *)cmd);
2480 match = iface_match(oif ? oif :
2481 m->m_pkthdr.rcvif, (ipfw_insn_if *)cmd);
2485 if (args->eh != NULL) { /* have MAC header */
2486 u_int32_t *want = (u_int32_t *)
2487 ((ipfw_insn_mac *)cmd)->addr;
2488 u_int32_t *mask = (u_int32_t *)
2489 ((ipfw_insn_mac *)cmd)->mask;
2490 u_int32_t *hdr = (u_int32_t *)args->eh;
2493 ( want[0] == (hdr[0] & mask[0]) &&
2494 want[1] == (hdr[1] & mask[1]) &&
2495 want[2] == (hdr[2] & mask[2]) );
2500 if (args->eh != NULL) {
2502 ntohs(args->eh->ether_type);
2504 ((ipfw_insn_u16 *)cmd)->ports;
2507 for (i = cmdlen - 1; !match && i>0;
2509 match = (t>=p[0] && t<=p[1]);
2514 match = (offset != 0);
2517 case O_IN: /* "out" is "not in" */
2518 match = (oif == NULL);
2522 match = (args->eh != NULL);
2526 match = (cmd->arg1 & 1 && divinput_flags &
2527 IP_FW_DIVERT_LOOPBACK_FLAG) ||
2528 (cmd->arg1 & 2 && divinput_flags &
2529 IP_FW_DIVERT_OUTPUT_FLAG);
2534 * We do not allow an arg of 0 so the
2535 * check of "proto" only suffices.
2537 match = (proto == cmd->arg1);
2542 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2546 case O_IP_SRC_LOOKUP:
2547 case O_IP_DST_LOOKUP:
2550 (cmd->opcode == O_IP_DST_LOOKUP) ?
2551 dst_ip.s_addr : src_ip.s_addr;
2554 match = lookup_table(chain, cmd->arg1, a,
2558 if (cmdlen == F_INSN_SIZE(ipfw_insn_u32))
2560 ((ipfw_insn_u32 *)cmd)->d[0] == v;
2570 (cmd->opcode == O_IP_DST_MASK) ?
2571 dst_ip.s_addr : src_ip.s_addr;
2572 uint32_t *p = ((ipfw_insn_u32 *)cmd)->d;
2575 for (; !match && i>0; i-= 2, p+= 2)
2576 match = (p[0] == (a & p[1]));
2584 INADDR_TO_IFP(src_ip, tif);
2585 match = (tif != NULL);
2592 u_int32_t *d = (u_int32_t *)(cmd+1);
2594 cmd->opcode == O_IP_DST_SET ?
2600 addr -= d[0]; /* subtract base */
2601 match = (addr < cmd->arg1) &&
2602 ( d[ 1 + (addr>>5)] &
2603 (1<<(addr & 0x1f)) );
2609 (((ipfw_insn_ip *)cmd)->addr.s_addr ==
2617 INADDR_TO_IFP(dst_ip, tif);
2618 match = (tif != NULL);
2625 * offset == 0 && proto != 0 is enough
2626 * to guarantee that we have a
2627 * packet with port info.
2629 if ((proto==IPPROTO_UDP || proto==IPPROTO_TCP)
2632 (cmd->opcode == O_IP_SRCPORT) ?
2633 src_port : dst_port ;
2635 ((ipfw_insn_u16 *)cmd)->ports;
2638 for (i = cmdlen - 1; !match && i>0;
2640 match = (x>=p[0] && x<=p[1]);
2645 match = (offset == 0 && proto==IPPROTO_ICMP &&
2646 icmptype_match(ICMP(ulp), (ipfw_insn_u32 *)cmd) );
2651 match = is_ipv6 && offset == 0 &&
2652 proto==IPPROTO_ICMPV6 &&
2654 ICMP6(ulp)->icmp6_type,
2655 (ipfw_insn_u32 *)cmd);
2661 ipopts_match(mtod(m, struct ip *), cmd) );
2666 cmd->arg1 == mtod(m, struct ip *)->ip_v);
2672 if (is_ipv4) { /* only for IP packets */
2677 if (cmd->opcode == O_IPLEN)
2679 else if (cmd->opcode == O_IPTTL)
2680 x = mtod(m, struct ip *)->ip_ttl;
2681 else /* must be IPID */
2682 x = ntohs(mtod(m, struct ip *)->ip_id);
2684 match = (cmd->arg1 == x);
2687 /* otherwise we have ranges */
2688 p = ((ipfw_insn_u16 *)cmd)->ports;
2690 for (; !match && i>0; i--, p += 2)
2691 match = (x >= p[0] && x <= p[1]);
2695 case O_IPPRECEDENCE:
2697 (cmd->arg1 == (mtod(m, struct ip *)->ip_tos & 0xe0)) );
2702 flags_match(cmd, mtod(m, struct ip *)->ip_tos));
2706 if (proto == IPPROTO_TCP && offset == 0) {
2714 ((ip->ip_hl + tcp->th_off) << 2);
2716 match = (cmd->arg1 == x);
2719 /* otherwise we have ranges */
2720 p = ((ipfw_insn_u16 *)cmd)->ports;
2722 for (; !match && i>0; i--, p += 2)
2723 match = (x >= p[0] && x <= p[1]);
2728 match = (proto == IPPROTO_TCP && offset == 0 &&
2729 flags_match(cmd, TCP(ulp)->th_flags));
2733 match = (proto == IPPROTO_TCP && offset == 0 &&
2734 tcpopts_match(TCP(ulp), cmd));
2738 match = (proto == IPPROTO_TCP && offset == 0 &&
2739 ((ipfw_insn_u32 *)cmd)->d[0] ==
2744 match = (proto == IPPROTO_TCP && offset == 0 &&
2745 ((ipfw_insn_u32 *)cmd)->d[0] ==
2750 match = (proto == IPPROTO_TCP && offset == 0 &&
2751 cmd->arg1 == TCP(ulp)->th_win);
2755 /* reject packets which have SYN only */
2756 /* XXX should i also check for TH_ACK ? */
2757 match = (proto == IPPROTO_TCP && offset == 0 &&
2758 (TCP(ulp)->th_flags &
2759 (TH_RST | TH_ACK | TH_SYN)) != TH_SYN);
2763 struct altq_tag *at;
2764 ipfw_insn_altq *altq = (ipfw_insn_altq *)cmd;
2767 mtag = m_tag_find(m, PACKET_TAG_PF_QID, NULL);
2770 mtag = m_tag_get(PACKET_TAG_PF_QID,
2771 sizeof(struct altq_tag),
2775 * Let the packet fall back to the
2780 at = (struct altq_tag *)(mtag+1);
2781 at->qid = altq->qid;
2787 m_tag_prepend(m, mtag);
2793 ipfw_log(f, hlen, args, m, oif, offset);
2798 match = (random()<((ipfw_insn_u32 *)cmd)->d[0]);
2802 /* Outgoing packets automatically pass/match */
2803 match = ((oif != NULL) ||
2804 (m->m_pkthdr.rcvif == NULL) ||
2808 verify_path6(&(args->f_id.src_ip6),
2809 m->m_pkthdr.rcvif) :
2811 verify_path(src_ip, m->m_pkthdr.rcvif)));
2815 /* Outgoing packets automatically pass/match */
2816 match = (hlen > 0 && ((oif != NULL) ||
2819 verify_path6(&(args->f_id.src_ip6),
2822 verify_path(src_ip, NULL)));
2826 /* Outgoing packets automatically pass/match */
2827 if (oif == NULL && hlen > 0 &&
2828 ( (is_ipv4 && in_localaddr(src_ip))
2831 in6_localaddr(&(args->f_id.src_ip6)))
2836 is_ipv6 ? verify_path6(
2837 &(args->f_id.src_ip6),
2838 m->m_pkthdr.rcvif) :
2848 match = (m_tag_find(m,
2849 PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL);
2852 match = (ipsec_getnhist(m) != 0);
2854 /* otherwise no match */
2860 IN6_ARE_ADDR_EQUAL(&args->f_id.src_ip6,
2861 &((ipfw_insn_ip6 *)cmd)->addr6);
2866 IN6_ARE_ADDR_EQUAL(&args->f_id.dst_ip6,
2867 &((ipfw_insn_ip6 *)cmd)->addr6);
2869 case O_IP6_SRC_MASK:
2871 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2872 struct in6_addr p = args->f_id.src_ip6;
2874 APPLY_MASK(&p, &te->mask6);
2875 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2879 case O_IP6_DST_MASK:
2881 ipfw_insn_ip6 *te = (ipfw_insn_ip6 *)cmd;
2882 struct in6_addr p = args->f_id.dst_ip6;
2884 APPLY_MASK(&p, &te->mask6);
2885 match = IN6_ARE_ADDR_EQUAL(&te->addr6, &p);
2890 match= is_ipv6 && search_ip6_addr_net(&args->f_id.src_ip6);
2894 match= is_ipv6 && search_ip6_addr_net(&args->f_id.dst_ip6);
2899 flow6id_match(args->f_id.flow_id6,
2900 (ipfw_insn_u32 *) cmd);
2905 (ext_hd & ((ipfw_insn *) cmd)->arg1);
2918 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2919 tablearg : cmd->arg1;
2921 /* Packet is already tagged with this tag? */
2922 mtag = m_tag_locate(m, MTAG_IPFW, tag, NULL);
2924 /* We have `untag' action when F_NOT flag is
2925 * present. And we must remove this mtag from
2926 * mbuf and reset `match' to zero (`match' will
2927 * be inversed later).
2928 * Otherwise we should allocate new mtag and
2929 * push it into mbuf.
2931 if (cmd->len & F_NOT) { /* `untag' action */
2933 m_tag_delete(m, mtag);
2934 } else if (mtag == NULL) {
2935 if ((mtag = m_tag_alloc(MTAG_IPFW,
2936 tag, 0, M_NOWAIT)) != NULL)
2937 m_tag_prepend(m, mtag);
2939 match = (cmd->len & F_NOT) ? 0: 1;
2944 uint32_t tag = (cmd->arg1 == IP_FW_TABLEARG) ?
2945 tablearg : cmd->arg1;
2948 match = m_tag_locate(m, MTAG_IPFW,
2953 /* we have ranges */
2954 for (mtag = m_tag_first(m);
2955 mtag != NULL && !match;
2956 mtag = m_tag_next(m, mtag)) {
2960 if (mtag->m_tag_cookie != MTAG_IPFW)
2963 p = ((ipfw_insn_u16 *)cmd)->ports;
2965 for(; !match && i > 0; i--, p += 2)
2967 mtag->m_tag_id >= p[0] &&
2968 mtag->m_tag_id <= p[1];
2974 * The second set of opcodes represents 'actions',
2975 * i.e. the terminal part of a rule once the packet
2976 * matches all previous patterns.
2977 * Typically there is only one action for each rule,
2978 * and the opcode is stored at the end of the rule
2979 * (but there are exceptions -- see below).
2981 * In general, here we set retval and terminate the
2982 * outer loop (would be a 'break 3' in some language,
2983 * but we need to do a 'goto done').
2986 * O_COUNT and O_SKIPTO actions:
2987 * instead of terminating, we jump to the next rule
2988 * ('goto next_rule', equivalent to a 'break 2'),
2989 * or to the SKIPTO target ('goto again' after
2990 * having set f, cmd and l), respectively.
2992 * O_TAG, O_LOG and O_ALTQ action parameters:
2993 * perform some action and set match = 1;
2995 * O_LIMIT and O_KEEP_STATE: these opcodes are
2996 * not real 'actions', and are stored right
2997 * before the 'action' part of the rule.
2998 * These opcodes try to install an entry in the
2999 * state tables; if successful, we continue with
3000 * the next opcode (match=1; break;), otherwise
3001 * the packet * must be dropped
3002 * ('goto done' after setting retval);
3004 * O_PROBE_STATE and O_CHECK_STATE: these opcodes
3005 * cause a lookup of the state table, and a jump
3006 * to the 'action' part of the parent rule
3007 * ('goto check_body') if an entry is found, or
3008 * (CHECK_STATE only) a jump to the next rule if
3009 * the entry is not found ('goto next_rule').
3010 * The result of the lookup is cached to make
3011 * further instances of these opcodes are
3016 if (install_state(f,
3017 (ipfw_insn_limit *)cmd, args, tablearg)) {
3018 retval = IP_FW_DENY;
3019 goto done; /* error/limit violation */
3027 * dynamic rules are checked at the first
3028 * keep-state or check-state occurrence,
3029 * with the result being stored in dyn_dir.
3030 * The compiler introduces a PROBE_STATE
3031 * instruction for us when we have a
3032 * KEEP_STATE (because PROBE_STATE needs
3035 if (dyn_dir == MATCH_UNKNOWN &&
3036 (q = lookup_dyn_rule(&args->f_id,
3037 &dyn_dir, proto == IPPROTO_TCP ?
3041 * Found dynamic entry, update stats
3042 * and jump to the 'action' part of
3048 cmd = ACTION_PTR(f);
3049 l = f->cmd_len - f->act_ofs;
3054 * Dynamic entry not found. If CHECK_STATE,
3055 * skip to next rule, if PROBE_STATE just
3056 * ignore and continue with next opcode.
3058 if (cmd->opcode == O_CHECK_STATE)
3064 retval = 0; /* accept */
3069 args->rule = f; /* report matching rule */
3070 if (cmd->arg1 == IP_FW_TABLEARG)
3071 args->cookie = tablearg;
3073 args->cookie = cmd->arg1;
3074 retval = IP_FW_DUMMYNET;
3079 struct divert_tag *dt;
3081 if (args->eh) /* not on layer 2 */
3083 mtag = m_tag_get(PACKET_TAG_DIVERT,
3084 sizeof(struct divert_tag),
3089 IPFW_RUNLOCK(chain);
3090 return (IP_FW_DENY);
3092 dt = (struct divert_tag *)(mtag+1);
3093 dt->cookie = f->rulenum;
3094 if (cmd->arg1 == IP_FW_TABLEARG)
3095 dt->info = tablearg;
3097 dt->info = cmd->arg1;
3098 m_tag_prepend(m, mtag);
3099 retval = (cmd->opcode == O_DIVERT) ?
3100 IP_FW_DIVERT : IP_FW_TEE;
3106 f->pcnt++; /* update stats */
3108 f->timestamp = time_uptime;
3109 if (cmd->opcode == O_COUNT)
3112 if (f->next_rule == NULL)
3113 lookup_next_rule(f);
3119 * Drop the packet and send a reject notice
3120 * if the packet is not ICMP (or is an ICMP
3121 * query), and it is not multicast/broadcast.
3123 if (hlen > 0 && is_ipv4 && offset == 0 &&
3124 (proto != IPPROTO_ICMP ||
3125 is_icmp_query(ICMP(ulp))) &&
3126 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3127 !IN_MULTICAST(ntohl(dst_ip.s_addr))) {
3128 send_reject(args, cmd->arg1, ip_len);
3134 if (hlen > 0 && is_ipv6 &&
3135 ((offset & IP6F_OFF_MASK) == 0) &&
3136 (proto != IPPROTO_ICMPV6 ||
3137 (is_icmp6_query(args->f_id.flags) == 1)) &&
3138 !(m->m_flags & (M_BCAST|M_MCAST)) &&
3139 !IN6_IS_ADDR_MULTICAST(&args->f_id.dst_ip6)) {
3140 send_reject6(args, cmd->arg1, hlen);
3146 retval = IP_FW_DENY;
3150 if (args->eh) /* not valid on layer2 pkts */
3152 if (!q || dyn_dir == MATCH_FORWARD)
3154 &((ipfw_insn_sa *)cmd)->sa;
3155 retval = IP_FW_PASS;
3160 args->rule = f; /* report matching rule */
3161 if (cmd->arg1 == IP_FW_TABLEARG)
3162 args->cookie = tablearg;
3164 args->cookie = cmd->arg1;
3165 retval = (cmd->opcode == O_NETGRAPH) ?
3166 IP_FW_NETGRAPH : IP_FW_NGTEE;
3170 panic("-- unknown opcode %d\n", cmd->opcode);
3171 } /* end of switch() on opcodes */
3173 if (cmd->len & F_NOT)
3177 if (cmd->len & F_OR)
3180 if (!(cmd->len & F_OR)) /* not an OR block, */
3181 break; /* try next rule */
3184 } /* end of inner for, scan opcodes */
3186 next_rule:; /* try next rule */
3188 } /* end of outer for, scan rules */
3189 printf("ipfw: ouch!, skip past end of rules, denying packet\n");
3190 IPFW_RUNLOCK(chain);
3191 return (IP_FW_DENY);
3194 /* Update statistics */
3197 f->timestamp = time_uptime;
3198 IPFW_RUNLOCK(chain);
3203 printf("ipfw: pullup failed\n");
3204 return (IP_FW_DENY);
3208 * When a rule is added/deleted, clear the next_rule pointers in all rules.
3209 * These will be reconstructed on the fly as packets are matched.
3212 flush_rule_ptrs(struct ip_fw_chain *chain)
3216 IPFW_WLOCK_ASSERT(chain);
3218 for (rule = chain->rules; rule; rule = rule->next)
3219 rule->next_rule = NULL;
3223 * Add a new rule to the list. Copy the rule into a malloc'ed area, then
3224 * possibly create a rule number and add the rule to the list.
3225 * Update the rule_number in the input struct so the caller knows it as well.
3228 add_rule(struct ip_fw_chain *chain, struct ip_fw *input_rule)
3230 struct ip_fw *rule, *f, *prev;
3231 int l = RULESIZE(input_rule);
3233 if (chain->rules == NULL && input_rule->rulenum != IPFW_DEFAULT_RULE)
3236 rule = malloc(l, M_IPFW, M_NOWAIT | M_ZERO);
3240 bcopy(input_rule, rule, l);
3243 rule->next_rule = NULL;
3247 rule->timestamp = 0;
3251 if (chain->rules == NULL) { /* default rule */
3252 chain->rules = rule;
3257 * If rulenum is 0, find highest numbered rule before the
3258 * default rule, and add autoinc_step
3260 if (autoinc_step < 1)
3262 else if (autoinc_step > 1000)
3263 autoinc_step = 1000;
3264 if (rule->rulenum == 0) {
3266 * locate the highest numbered rule before default
3268 for (f = chain->rules; f; f = f->next) {
3269 if (f->rulenum == IPFW_DEFAULT_RULE)
3271 rule->rulenum = f->rulenum;
3273 if (rule->rulenum < IPFW_DEFAULT_RULE - autoinc_step)
3274 rule->rulenum += autoinc_step;
3275 input_rule->rulenum = rule->rulenum;
3279 * Now insert the new rule in the right place in the sorted list.
3281 for (prev = NULL, f = chain->rules; f; prev = f, f = f->next) {
3282 if (f->rulenum > rule->rulenum) { /* found the location */
3286 } else { /* head insert */
3287 rule->next = chain->rules;
3288 chain->rules = rule;
3293 flush_rule_ptrs(chain);
3297 IPFW_WUNLOCK(chain);
3298 DEB(printf("ipfw: installed rule %d, static count now %d\n",
3299 rule->rulenum, static_count);)
3304 * Remove a static rule (including derived * dynamic rules)
3305 * and place it on the ``reap list'' for later reclamation.
3306 * The caller is in charge of clearing rule pointers to avoid
3307 * dangling pointers.
3308 * @return a pointer to the next entry.
3309 * Arguments are not checked, so they better be correct.
3311 static struct ip_fw *
3312 remove_rule(struct ip_fw_chain *chain, struct ip_fw *rule, struct ip_fw *prev)
3315 int l = RULESIZE(rule);
3317 IPFW_WLOCK_ASSERT(chain);
3321 remove_dyn_rule(rule, NULL /* force removal */);
3330 rule->next = chain->reap;
3337 * Reclaim storage associated with a list of rules. This is
3338 * typically the list created using remove_rule.
3341 reap_rules(struct ip_fw *head)
3345 while ((rule = head) != NULL) {
3347 if (DUMMYNET_LOADED)
3348 ip_dn_ruledel_ptr(rule);
3354 * Remove all rules from a chain (except rules in set RESVD_SET
3355 * unless kill_default = 1). The caller is responsible for
3356 * reclaiming storage for the rules left in chain->reap.
3359 free_chain(struct ip_fw_chain *chain, int kill_default)
3361 struct ip_fw *prev, *rule;
3363 IPFW_WLOCK_ASSERT(chain);
3365 flush_rule_ptrs(chain); /* more efficient to do outside the loop */
3366 for (prev = NULL, rule = chain->rules; rule ; )
3367 if (kill_default || rule->set != RESVD_SET)
3368 rule = remove_rule(chain, rule, prev);
3376 * Remove all rules with given number, and also do set manipulation.
3377 * Assumes chain != NULL && *chain != NULL.
3379 * The argument is an u_int32_t. The low 16 bit are the rule or set number,
3380 * the next 8 bits are the new set, the top 8 bits are the command:
3382 * 0 delete rules with given number
3383 * 1 delete rules with given set number
3384 * 2 move rules with given number to new set
3385 * 3 move rules with given set number to new set
3386 * 4 swap sets with given numbers
3389 del_entry(struct ip_fw_chain *chain, u_int32_t arg)
3391 struct ip_fw *prev = NULL, *rule;
3392 u_int16_t rulenum; /* rule or old_set */
3393 u_int8_t cmd, new_set;
3395 rulenum = arg & 0xffff;
3396 cmd = (arg >> 24) & 0xff;
3397 new_set = (arg >> 16) & 0xff;
3401 if (new_set > RESVD_SET)
3403 if (cmd == 0 || cmd == 2) {
3404 if (rulenum >= IPFW_DEFAULT_RULE)
3407 if (rulenum > RESVD_SET) /* old_set */
3412 rule = chain->rules;
3415 case 0: /* delete rules with given number */
3417 * locate first rule to delete
3419 for (; rule->rulenum < rulenum; prev = rule, rule = rule->next)
3421 if (rule->rulenum != rulenum) {
3422 IPFW_WUNLOCK(chain);
3427 * flush pointers outside the loop, then delete all matching
3428 * rules. prev remains the same throughout the cycle.
3430 flush_rule_ptrs(chain);
3431 while (rule->rulenum == rulenum)
3432 rule = remove_rule(chain, rule, prev);
3435 case 1: /* delete all rules with given set number */
3436 flush_rule_ptrs(chain);
3437 rule = chain->rules;
3438 while (rule->rulenum < IPFW_DEFAULT_RULE)
3439 if (rule->set == rulenum)
3440 rule = remove_rule(chain, rule, prev);
3447 case 2: /* move rules with given number to new set */
3448 rule = chain->rules;
3449 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3450 if (rule->rulenum == rulenum)
3451 rule->set = new_set;
3454 case 3: /* move rules with given set number to new set */
3455 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3456 if (rule->set == rulenum)
3457 rule->set = new_set;
3460 case 4: /* swap two sets */
3461 for (; rule->rulenum < IPFW_DEFAULT_RULE; rule = rule->next)
3462 if (rule->set == rulenum)
3463 rule->set = new_set;
3464 else if (rule->set == new_set)
3465 rule->set = rulenum;
3469 * Look for rules to reclaim. We grab the list before
3470 * releasing the lock then reclaim them w/o the lock to
3471 * avoid a LOR with dummynet.
3475 IPFW_WUNLOCK(chain);
3482 * Clear counters for a specific rule.
3483 * The enclosing "table" is assumed locked.
3486 clear_counters(struct ip_fw *rule, int log_only)
3488 ipfw_insn_log *l = (ipfw_insn_log *)ACTION_PTR(rule);
3490 if (log_only == 0) {
3491 rule->bcnt = rule->pcnt = 0;
3492 rule->timestamp = 0;
3494 if (l->o.opcode == O_LOG)
3495 l->log_left = l->max_log;
3499 * Reset some or all counters on firewall rules.
3500 * @arg frwl is null to clear all entries, or contains a specific
3502 * @arg log_only is 1 if we only want to reset logs, zero otherwise.
3505 zero_entry(struct ip_fw_chain *chain, int rulenum, int log_only)
3513 for (rule = chain->rules; rule; rule = rule->next)
3514 clear_counters(rule, log_only);
3515 msg = log_only ? "ipfw: All logging counts reset.\n" :
3516 "ipfw: Accounting cleared.\n";
3520 * We can have multiple rules with the same number, so we
3521 * need to clear them all.
3523 for (rule = chain->rules; rule; rule = rule->next)
3524 if (rule->rulenum == rulenum) {
3525 while (rule && rule->rulenum == rulenum) {
3526 clear_counters(rule, log_only);
3532 if (!cleared) { /* we did not find any matching rules */
3533 IPFW_WUNLOCK(chain);
3536 msg = log_only ? "ipfw: Entry %d logging count reset.\n" :
3537 "ipfw: Entry %d cleared.\n";
3539 IPFW_WUNLOCK(chain);
3542 log(LOG_SECURITY | LOG_NOTICE, msg, rulenum);
3547 * Check validity of the structure before insert.
3548 * Fortunately rules are simple, so this mostly need to check rule sizes.
3551 check_ipfw_struct(struct ip_fw *rule, int size)
3557 if (size < sizeof(*rule)) {
3558 printf("ipfw: rule too short\n");
3561 /* first, check for valid size */
3564 printf("ipfw: size mismatch (have %d want %d)\n", size, l);
3567 if (rule->act_ofs >= rule->cmd_len) {
3568 printf("ipfw: bogus action offset (%u > %u)\n",
3569 rule->act_ofs, rule->cmd_len - 1);
3573 * Now go for the individual checks. Very simple ones, basically only
3574 * instruction sizes.
3576 for (l = rule->cmd_len, cmd = rule->cmd ;
3577 l > 0 ; l -= cmdlen, cmd += cmdlen) {
3578 cmdlen = F_LEN(cmd);
3580 printf("ipfw: opcode %d size truncated\n",
3584 DEB(printf("ipfw: opcode %d\n", cmd->opcode);)
3585 switch (cmd->opcode) {
3597 case O_IPPRECEDENCE:
3615 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3628 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3633 if (cmdlen != F_INSN_SIZE(ipfw_insn_limit))
3638 if (cmdlen != F_INSN_SIZE(ipfw_insn_log))
3641 ((ipfw_insn_log *)cmd)->log_left =
3642 ((ipfw_insn_log *)cmd)->max_log;
3648 /* only odd command lengths */
3649 if ( !(cmdlen & 1) || cmdlen > 31)
3655 if (cmd->arg1 == 0 || cmd->arg1 > 256) {
3656 printf("ipfw: invalid set size %d\n",
3660 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3665 case O_IP_SRC_LOOKUP:
3666 case O_IP_DST_LOOKUP:
3667 if (cmd->arg1 >= IPFW_TABLES_MAX) {
3668 printf("ipfw: invalid table number %d\n",
3672 if (cmdlen != F_INSN_SIZE(ipfw_insn) &&
3673 cmdlen != F_INSN_SIZE(ipfw_insn_u32))
3678 if (cmdlen != F_INSN_SIZE(ipfw_insn_mac))
3688 if (cmdlen < 1 || cmdlen > 31)
3694 case O_IP_DSTPORT: /* XXX artificial limit, 30 port pairs */
3695 if (cmdlen < 2 || cmdlen > 31)
3702 if (cmdlen != F_INSN_SIZE(ipfw_insn_if))
3707 if (cmdlen != F_INSN_SIZE(ipfw_insn_altq))
3713 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3718 #ifdef IPFIREWALL_FORWARD
3719 if (cmdlen != F_INSN_SIZE(ipfw_insn_sa))
3728 if (ip_divert_ptr == NULL)
3734 if (!NG_IPFW_LOADED)
3738 case O_FORWARD_MAC: /* XXX not implemented yet */
3749 if (cmdlen != F_INSN_SIZE(ipfw_insn))
3753 printf("ipfw: opcode %d, multiple actions"
3760 printf("ipfw: opcode %d, action must be"
3769 if (cmdlen != F_INSN_SIZE(struct in6_addr) +
3770 F_INSN_SIZE(ipfw_insn))
3775 if (cmdlen != F_INSN_SIZE(ipfw_insn_u32) +
3776 ((ipfw_insn_u32 *)cmd)->o.arg1)
3780 case O_IP6_SRC_MASK:
3781 case O_IP6_DST_MASK:
3782 if ( !(cmdlen & 1) || cmdlen > 127)
3786 if( cmdlen != F_INSN_SIZE( ipfw_insn_icmp6 ) )
3792 switch (cmd->opcode) {
3802 case O_IP6_SRC_MASK:
3803 case O_IP6_DST_MASK:
3805 printf("ipfw: no IPv6 support in kernel\n");
3806 return EPROTONOSUPPORT;
3809 printf("ipfw: opcode %d, unknown opcode\n",
3815 if (have_action == 0) {
3816 printf("ipfw: missing action\n");
3822 printf("ipfw: opcode %d size %d wrong\n",
3823 cmd->opcode, cmdlen);
3828 * Copy the static and dynamic rules to the supplied buffer
3829 * and return the amount of space actually used.
3832 ipfw_getrules(struct ip_fw_chain *chain, void *buf, size_t space)
3835 char *ep = bp + space;
3839 /* XXX this can take a long time and locking will block packet flow */
3841 for (rule = chain->rules; rule ; rule = rule->next) {
3843 * Verify the entry fits in the buffer in case the
3844 * rules changed between calculating buffer space and
3845 * now. This would be better done using a generation
3846 * number but should suffice for now.
3851 bcopy(&set_disable, &(((struct ip_fw *)bp)->next_rule),
3852 sizeof(set_disable));
3856 IPFW_RUNLOCK(chain);
3858 ipfw_dyn_rule *p, *last = NULL;
3861 for (i = 0 ; i < curr_dyn_buckets; i++)
3862 for (p = ipfw_dyn_v[i] ; p != NULL; p = p->next) {
3863 if (bp + sizeof *p <= ep) {
3864 ipfw_dyn_rule *dst =
3865 (ipfw_dyn_rule *)bp;
3866 bcopy(p, dst, sizeof *p);
3867 bcopy(&(p->rule->rulenum), &(dst->rule),
3868 sizeof(p->rule->rulenum));
3870 * store a non-null value in "next".
3871 * The userland code will interpret a
3872 * NULL here as a marker
3873 * for the last dynamic rule.
3875 bcopy(&dst, &dst->next, sizeof(dst));
3878 TIME_LEQ(dst->expire, time_uptime) ?
3879 0 : dst->expire - time_uptime ;
3880 bp += sizeof(ipfw_dyn_rule);
3884 if (last != NULL) /* mark last dynamic rule */
3885 bzero(&last->next, sizeof(last));
3887 return (bp - (char *)buf);
3892 * {set|get}sockopt parser.
3895 ipfw_ctl(struct sockopt *sopt)
3897 #define RULE_MAXSIZE (256*sizeof(u_int32_t))
3898 int error, rule_num;
3900 struct ip_fw *buf, *rule;
3901 u_int32_t rulenum[2];
3903 error = suser(sopt->sopt_td);
3908 * Disallow modifications in really-really secure mode, but still allow
3909 * the logging counters to be reset.
3911 if (sopt->sopt_name == IP_FW_ADD ||
3912 (sopt->sopt_dir == SOPT_SET && sopt->sopt_name != IP_FW_RESETLOG)) {
3913 error = securelevel_ge(sopt->sopt_td->td_ucred, 3);
3920 switch (sopt->sopt_name) {
3923 * pass up a copy of the current rules. Static rules
3924 * come first (the last of which has number IPFW_DEFAULT_RULE),
3925 * followed by a possibly empty list of dynamic rule.
3926 * The last dynamic rule has NULL in the "next" field.
3928 * Note that the calculated size is used to bound the
3929 * amount of data returned to the user. The rule set may
3930 * change between calculating the size and returning the
3931 * data in which case we'll just return what fits.
3933 size = static_len; /* size of static rules */
3934 if (ipfw_dyn_v) /* add size of dyn.rules */
3935 size += (dyn_count * sizeof(ipfw_dyn_rule));
3938 * XXX todo: if the user passes a short length just to know
3939 * how much room is needed, do not bother filling up the
3940 * buffer, just jump to the sooptcopyout.
3942 buf = malloc(size, M_TEMP, M_WAITOK);
3943 error = sooptcopyout(sopt, buf,
3944 ipfw_getrules(&layer3_chain, buf, size));
3950 * Normally we cannot release the lock on each iteration.
3951 * We could do it here only because we start from the head all
3952 * the times so there is no risk of missing some entries.
3953 * On the other hand, the risk is that we end up with
3954 * a very inconsistent ruleset, so better keep the lock
3955 * around the whole cycle.
3957 * XXX this code can be improved by resetting the head of
3958 * the list to point to the default rule, and then freeing
3959 * the old list without the need for a lock.
3962 IPFW_WLOCK(&layer3_chain);
3963 layer3_chain.reap = NULL;
3964 free_chain(&layer3_chain, 0 /* keep default rule */);
3965 rule = layer3_chain.reap, layer3_chain.reap = NULL;
3966 IPFW_WUNLOCK(&layer3_chain);
3967 if (layer3_chain.reap != NULL)
3972 rule = malloc(RULE_MAXSIZE, M_TEMP, M_WAITOK);
3973 error = sooptcopyin(sopt, rule, RULE_MAXSIZE,
3974 sizeof(struct ip_fw) );
3976 error = check_ipfw_struct(rule, sopt->sopt_valsize);
3978 error = add_rule(&layer3_chain, rule);
3979 size = RULESIZE(rule);
3980 if (!error && sopt->sopt_dir == SOPT_GET)
3981 error = sooptcopyout(sopt, rule, size);
3988 * IP_FW_DEL is used for deleting single rules or sets,
3989 * and (ab)used to atomically manipulate sets. Argument size
3990 * is used to distinguish between the two:
3992 * delete single rule or set of rules,
3993 * or reassign rules (or sets) to a different set.
3994 * 2*sizeof(u_int32_t)
3995 * atomic disable/enable sets.
3996 * first u_int32_t contains sets to be disabled,
3997 * second u_int32_t contains sets to be enabled.
3999 error = sooptcopyin(sopt, rulenum,
4000 2*sizeof(u_int32_t), sizeof(u_int32_t));
4003 size = sopt->sopt_valsize;
4004 if (size == sizeof(u_int32_t)) /* delete or reassign */
4005 error = del_entry(&layer3_chain, rulenum[0]);
4006 else if (size == 2*sizeof(u_int32_t)) /* set enable/disable */
4008 (set_disable | rulenum[0]) & ~rulenum[1] &
4009 ~(1<<RESVD_SET); /* set RESVD_SET always enabled */
4015 case IP_FW_RESETLOG: /* argument is an int, the rule number */
4017 if (sopt->sopt_val != 0) {
4018 error = sooptcopyin(sopt, &rule_num,
4019 sizeof(int), sizeof(int));
4023 error = zero_entry(&layer3_chain, rule_num,
4024 sopt->sopt_name == IP_FW_RESETLOG);
4027 case IP_FW_TABLE_ADD:
4029 ipfw_table_entry ent;
4031 error = sooptcopyin(sopt, &ent,
4032 sizeof(ent), sizeof(ent));
4035 error = add_table_entry(&layer3_chain, ent.tbl,
4036 ent.addr, ent.masklen, ent.value);
4040 case IP_FW_TABLE_DEL:
4042 ipfw_table_entry ent;
4044 error = sooptcopyin(sopt, &ent,
4045 sizeof(ent), sizeof(ent));
4048 error = del_table_entry(&layer3_chain, ent.tbl,
4049 ent.addr, ent.masklen);
4053 case IP_FW_TABLE_FLUSH:
4057 error = sooptcopyin(sopt, &tbl,
4058 sizeof(tbl), sizeof(tbl));
4061 IPFW_WLOCK(&layer3_chain);
4062 error = flush_table(&layer3_chain, tbl);
4063 IPFW_WUNLOCK(&layer3_chain);
4067 case IP_FW_TABLE_GETSIZE:
4071 if ((error = sooptcopyin(sopt, &tbl, sizeof(tbl),
4074 IPFW_RLOCK(&layer3_chain);
4075 error = count_table(&layer3_chain, tbl, &cnt);
4076 IPFW_RUNLOCK(&layer3_chain);
4079 error = sooptcopyout(sopt, &cnt, sizeof(cnt));
4083 case IP_FW_TABLE_LIST:
4087 if (sopt->sopt_valsize < sizeof(*tbl)) {
4091 size = sopt->sopt_valsize;
4092 tbl = malloc(size, M_TEMP, M_WAITOK);
4097 error = sooptcopyin(sopt, tbl, size, sizeof(*tbl));
4102 tbl->size = (size - sizeof(*tbl)) /
4103 sizeof(ipfw_table_entry);
4104 IPFW_RLOCK(&layer3_chain);
4105 error = dump_table(&layer3_chain, tbl);
4106 IPFW_RUNLOCK(&layer3_chain);
4111 error = sooptcopyout(sopt, tbl, size);
4117 printf("ipfw: ipfw_ctl invalid option %d\n", sopt->sopt_name);
4126 * dummynet needs a reference to the default rule, because rules can be
4127 * deleted while packets hold a reference to them. When this happens,
4128 * dummynet changes the reference to the default rule (it could well be a
4129 * NULL pointer, but this way we do not need to check for the special
4130 * case, plus here he have info on the default behaviour).
4132 struct ip_fw *ip_fw_default_rule;
4135 * This procedure is only used to handle keepalives. It is invoked
4136 * every dyn_keepalive_period
4139 ipfw_tick(void * __unused unused)
4141 struct mbuf *m0, *m, *mnext, **mtailp;
4145 if (dyn_keepalive == 0 || ipfw_dyn_v == NULL || dyn_count == 0)
4149 * We make a chain of packets to go out here -- not deferring
4150 * until after we drop the IPFW dynamic rule lock would result
4151 * in a lock order reversal with the normal packet input -> ipfw
4157 for (i = 0 ; i < curr_dyn_buckets ; i++) {
4158 for (q = ipfw_dyn_v[i] ; q ; q = q->next ) {
4159 if (q->dyn_type == O_LIMIT_PARENT)
4161 if (q->id.proto != IPPROTO_TCP)
4163 if ( (q->state & BOTH_SYN) != BOTH_SYN)
4165 if (TIME_LEQ( time_uptime+dyn_keepalive_interval,
4167 continue; /* too early */
4168 if (TIME_LEQ(q->expire, time_uptime))
4169 continue; /* too late, rule expired */
4171 *mtailp = send_pkt(&(q->id), q->ack_rev - 1,
4172 q->ack_fwd, TH_SYN);
4173 if (*mtailp != NULL)
4174 mtailp = &(*mtailp)->m_nextpkt;
4175 *mtailp = send_pkt(&(q->id), q->ack_fwd - 1,
4177 if (*mtailp != NULL)
4178 mtailp = &(*mtailp)->m_nextpkt;
4182 for (m = mnext = m0; m != NULL; m = mnext) {
4183 mnext = m->m_nextpkt;
4184 m->m_nextpkt = NULL;
4185 ip_output(m, NULL, NULL, 0, NULL, NULL);
4188 callout_reset(&ipfw_timeout, dyn_keepalive_period*hz, ipfw_tick, NULL);
4194 struct ip_fw default_rule;
4198 /* Setup IPv6 fw sysctl tree. */
4199 sysctl_ctx_init(&ip6_fw_sysctl_ctx);
4200 ip6_fw_sysctl_tree = SYSCTL_ADD_NODE(&ip6_fw_sysctl_ctx,
4201 SYSCTL_STATIC_CHILDREN(_net_inet6_ip6), OID_AUTO, "fw",
4202 CTLFLAG_RW | CTLFLAG_SECURE, 0, "Firewall");
4203 SYSCTL_ADD_PROC(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4204 OID_AUTO, "enable", CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_SECURE3,
4205 &fw6_enable, 0, ipfw_chg_hook, "I", "Enable ipfw+6");
4206 SYSCTL_ADD_INT(&ip6_fw_sysctl_ctx, SYSCTL_CHILDREN(ip6_fw_sysctl_tree),
4207 OID_AUTO, "deny_unknown_exthdrs", CTLFLAG_RW | CTLFLAG_SECURE,
4208 &fw_deny_unknown_exthdrs, 0,
4209 "Deny packets with unknown IPv6 Extension Headers");
4212 layer3_chain.rules = NULL;
4213 IPFW_LOCK_INIT(&layer3_chain);
4214 ipfw_dyn_rule_zone = uma_zcreate("IPFW dynamic rule zone",
4215 sizeof(ipfw_dyn_rule), NULL, NULL, NULL, NULL,
4217 IPFW_DYN_LOCK_INIT();
4218 callout_init(&ipfw_timeout, NET_CALLOUT_MPSAFE);
4220 bzero(&default_rule, sizeof default_rule);
4222 default_rule.act_ofs = 0;
4223 default_rule.rulenum = IPFW_DEFAULT_RULE;
4224 default_rule.cmd_len = 1;
4225 default_rule.set = RESVD_SET;
4227 default_rule.cmd[0].len = 1;
4228 default_rule.cmd[0].opcode =
4229 #ifdef IPFIREWALL_DEFAULT_TO_ACCEPT
4234 error = add_rule(&layer3_chain, &default_rule);
4236 printf("ipfw2: error %u initializing default rule "
4237 "(support disabled)\n", error);
4238 IPFW_DYN_LOCK_DESTROY();
4239 IPFW_LOCK_DESTROY(&layer3_chain);
4240 uma_zdestroy(ipfw_dyn_rule_zone);
4244 ip_fw_default_rule = layer3_chain.rules;
4249 "initialized, divert %s, "
4250 "rule-based forwarding "
4251 #ifdef IPFIREWALL_FORWARD
4256 "default to %s, logging ",
4262 default_rule.cmd[0].opcode == O_ACCEPT ? "accept" : "deny");
4264 #ifdef IPFIREWALL_VERBOSE
4267 #ifdef IPFIREWALL_VERBOSE_LIMIT
4268 verbose_limit = IPFIREWALL_VERBOSE_LIMIT;
4270 if (fw_verbose == 0)
4271 printf("disabled\n");
4272 else if (verbose_limit == 0)
4273 printf("unlimited\n");
4275 printf("limited to %d packets/entry by default\n",
4278 error = init_tables(&layer3_chain);
4280 IPFW_DYN_LOCK_DESTROY();
4281 IPFW_LOCK_DESTROY(&layer3_chain);
4282 uma_zdestroy(ipfw_dyn_rule_zone);
4285 ip_fw_ctl_ptr = ipfw_ctl;
4286 ip_fw_chk_ptr = ipfw_chk;
4287 callout_reset(&ipfw_timeout, hz, ipfw_tick, NULL);
4297 ip_fw_chk_ptr = NULL;
4298 ip_fw_ctl_ptr = NULL;
4299 callout_drain(&ipfw_timeout);
4300 IPFW_WLOCK(&layer3_chain);
4301 flush_tables(&layer3_chain);
4302 layer3_chain.reap = NULL;
4303 free_chain(&layer3_chain, 1 /* kill default rule */);
4304 reap = layer3_chain.reap, layer3_chain.reap = NULL;
4305 IPFW_WUNLOCK(&layer3_chain);
4308 IPFW_DYN_LOCK_DESTROY();
4309 uma_zdestroy(ipfw_dyn_rule_zone);
4310 IPFW_LOCK_DESTROY(&layer3_chain);
4313 /* Free IPv6 fw sysctl tree. */
4314 sysctl_ctx_free(&ip6_fw_sysctl_ctx);
4317 printf("IP firewall unloaded\n");